leibniz s syncategorematic infinitesimals smooth ...rarthur/articles/lsi-final.pdf · infinitesimal...
TRANSCRIPT
LEIBNIZ’S SYNCATEGOREMATIC INFINITESIMALS, SMOOTH INFINITESIMAL ANALYSIS AND SECOND-ORDER DIFFERENTIALS
Richard T. W. Arthur Department of Philosophy
McMaster University Hamilton, Ontario L8S 4K1
Canada
email: [email protected] phone: (905) 525-9140, x-23470
fax: (905) 577-0385 Abstract In contrast with some recent theories of infinitesimals as non-Archimedean entities, Leibniz’s mature interpretation was fully in accord with the Archimedean Axiom: infinitesimals are fictions, whose treatment as entities incomparably smaller than finite quantities is justifiable wholly in terms of variable finite quantities that can be taken as small as desired, i.e. syncategorematically. In this paper I explain this syncategorematic interpretation, and how Leibniz used it to justify the calculus. I then compare it with the approach of Smooth Infinitesimal Analysis, as propounded by John Bell. I find some salient differences, especially with regard to higher-order infinitesimals. I illustrate these differences by a consideration of how each approach might be applied to propositions of Newton’s Principia concerning the derivation of force laws for bodies orbiting in a circle and an ellipse.
Leibniz’s Syncategorematic Infinitesimals R. T. W. Arthur
1
“If the Leibnizian calculus needs a rehabilitation because of too severe treatment by historians in the past half century, as Robinson suggests (1966, 250), I feel that the legitimate grounds for such a rehabilitation are to be found in the Leibnizian theory itself.”—H. J. M. Bos (1974-75, 82-3)
1. INTRODUCTION
Leibniz’s doctrine of the fictional nature of infinitesimals has been much misunderstood. It
has been construed as a late defensive parry, an attempt to defend the success of his
infinitesimal calculus—understood as implicitly committed to the existence of infinitesimals as
actually infinitely small entities—in the face of criticisms of the adequacy of its foundations.
It has also been read as being at odds with other defences of the calculus Leibniz gave on
explicitly Archimedean foundations. I take the position here (following Ishiguro 1990) that the
idea that Leibniz was committed to infinitesimals as actually infinitely small entities is a
misreading: his mature interpretation of the calculus was fully in accord with the
Archimedean Axiom. Leibniz’s interpretation is (to use the medieval term) syncategorematic:
Infinitesimals are fictions in the sense that the terms designating them can be treated as if
they refer to entities incomparably smaller than finite quantities, but really stand for variable
finite quantities that can be taken as small as desired. As I (Arthur 2008) and Sam Levey
(2008) have argued, this interpretation is no late stratagem, but in place already by 1676.
In section 2 I present this interpretation by tracing its development from Leibniz’s
early work on infinite series and quadratures to his unpublished attempt in 1701 in the tract
“Cum prodiisset” to provide a systematic foundation for his calculus. We shall see that as
early as 1676 Leibniz had succeeded in providing a rigorous foundation for Riemannian
integration, based on the Archimedean Axiom. The appeal to this axiom, generalized by
Leibniz into his Law of Continuity, undergirds his interpretation of infinitesimals as fictions
that can nevertheless be used in calculations, and forms the basis for his foundation for
differentials in “Cum prodiisset”.
Leibniz’s Syncategorematic Infinitesimals R. T. W. Arthur
2
I then turn to a comparison of Leibniz’s approach with the recent theory of infinitesimals
championed by John Bell, Smooth Infinitesimal Analysis (SIA), of which I give a brief synopsis
in section 3. As we shall see, this has many points in common with Leibniz’s approach: the
non-punctiform nature of infinitesimals, their acting as parts of the continuum, the
dependence on variables (as opposed to the static quantities of both Standard and Non-
standard Analysis), the resolution of curves into infinite-sided polygons, and the finessing of a
commitment to the existence of infinitesimals. Nevertheless, there are also crucial
differences. These are brought into relief in sections 4 and 5, by a consideration of how each
approach might be applied to some of the central results in cosmology concerning force laws
for planetary motions. In section 4 I explain how Leibniz uses differential equations to re-
derive the results Newton had obtained for the v2/r law for the centripetal force on a body
orbiting around a centre of force and for the inverse square law. After briefly describing
Newton’s own derivations of Propositions 4, 6 and 11 in his Principia, I detail Leibniz’s
derivations of the “solicitation” ddr experienced by an orbiting body in a circular and then
elliptical orbits. Then in section 5 I consider the question of the legitimacy of Leibniz’s use of
second-order differentials, by reference to his own attempts to provide a foundation. It is
found that while Leibniz’s syncategorematic approach is adequate to ground his derivations
using second-order differentials, there is no corresponding possibility for a derivation of such
expressions using nilsquare infinitesimals; that is to say, while one can define derivatives of
all orders using SIA and thus effect proofs of all the propositions concerning orbiting bodies,
and while one can also define differentials of second and higher orders in SIA, the nilsquare
property precludes duplication of the calculations of Newton and Leibniz which involved dt2.
So not all the results obtained classically by the use of infinitesimals can be duplicated by SIA:
even if the infinitesimals of SIA are much closer in conception to Leibnizian infinitesimals
than those of Non-standard Analysis, they are not Leibnizian infinitesimals.
Leibniz’s Syncategorematic Infinitesimals R. T. W. Arthur
3
2. LEIBNIZ’S FOUNDATION FOR THE INFINITESIMAL CALCULUS
As Leibniz explains in his mature work, the infinite should not be understood to refer to an
actual entity that is greater than any finite one of the same kind, i.e. categorematically, but
rather syncategorematically: in certain well-defined circumstances infinite terms may be
used as if they refer to entities incomparably larger than finite quantities, but really stand for
variable finite quantities that can be taken as large as desired. Leibniz’s interpretation of
infinitesimals as fictions is intimately linked with his doctrine of the infinite. Just as the
infinite is not an actually existing whole made up of finite parts, so infinitesimals are not
actually existing parts which can be composed into a finite whole. As Leibniz explained to Des
Bosses at the beginning of their correspondence in 1706,
Speaking philosophically, I maintain that there are no more infinitely small magnitudes
than there are infinitely large ones, that is, no more infinitesimals than infinituples.
For I hold both to be fictions of the mind through an abbreviated way of speaking,
adapted to calculation, as imaginary roots in algebra are too. Meanwhile I have
demonstrated that these expressions are very useful for abbreviating thought and thus
for discovery, and cannot lead to error, since it suffices to substitute for the infinitely
small something as small as one wishes, so that the error is smaller than any given,
whence it follows that there can be no error. (GP II 305)
The roots of this syncategorematic interpretation, like the roots of Leibniz’s calculus itself,
can be discerned in his earliest work on quadratures and infinite series, specifically in his
work on the hyperbola. Already in the Fall of 1672 in his reading notes on Galileo Galilei’s
Two New Sciences (Discorsi), Leibniz had formulated the basis of his later position. There he
notes Galileo’s demonstration that there are as many square roots of (natural) numbers as
Leibniz’s Syncategorematic Infinitesimals R. T. W. Arthur
4
there are natural numbers, and that “therefore there are as many squares as numbers; which
is impossible”,1 and then comments:
Hence it follows either that in the infinite the whole is not greater than the part,
which is the opinion of Galileo and Gregory of St. Vincent,2 and which I cannot accept;
or that infinity itself is nothing, i.e. that it is not one and not a whole. Or perhaps we
should say: when distinguishing between infinities, the most infinite, i.e. all the
numbers, is something that implies a contradiction, for if it were a whole it could be
understood as made up of all the numbers continuing to infinity, and would be much
greater than all the numbers, that is, greater than the greatest number. Or perhaps
we should say that one ought not to say anything about the infinite, as a whole,
except when there is a demonstration of it. (A VI iii, 168; LoC 8-9)
Two features of this commentary are seminal for Leibniz’s later thought: the upholding of the
part-whole axiom for the infinite, with the entailed denial that the infinite is a whole; and
the more nuanced claim that one cannot assert a property of the infinite except insofar as
one has an independent demonstration of it.
The first claim is graphically illustrated by a calculation Leibniz performs in 1674, in
which we are confronted with an infinite whole which has a direct visual representation: the
area under a hyperbola between 0 and 1. By this time, Leibniz has made great strides in
summing infinite series, and applying the results to calculate “quadratures”, the areas under
curves. In this example, he uses his knowledge to demonstrate the inapplicability of the part-
whole axiom to infinite quantities.
1 Relevant extracts from Galileo’s Discorsi may be found in (LoC 352-357). 2 At EN 78, Salviati says: “for I believe that these attributes of greatness, smallness and equality do not befit infinities, of which one cannot be said to be greater than, smaller than, or equal to another”; and again at the end of the ensuing proof: “in final conclusion, the attributes of equal, greater, and less have no place in infinities, but only in bounded quantities” (EN 79; LoC 355-56). For Gregory of St Vincent's opinion, see his 1647, lib. 8, pr. 1, theorema, 870 ff.
Leibniz’s Syncategorematic Infinitesimals R. T. W. Arthur
5
Figure 1: Leibniz’s hyperbola
A C F
M
D G
E
C B
H L
F G
M
In the above diagram, which Leibniz has deliberately drawn symmetrically, MGBEM is a
hyperbola, and the variable abscissa DE = x (dashed line) is taken to vary between CB (taken
as the x-axis) and (the horizontal) ACF… . Leibniz calculates the area by “applying” DE to (the
vertical) AC (= 1) as a base. He expands DE as (1 – y)–1 = 1 + y + y2 + y3 + ..., and gets the
result ACBEM = 1 + 1/2 + 1/3 + 1/4 + 1/5 + … In modern terms, Leibniz is evaluating the area
under the curve x = (1 – y)–1 by calculating the definite integral ∫01 x dy using a series
expansion, to obtain
ACBEM = ∫01(1 – y)–1 dy = [y + y2/2 + y3/3 + y4/4... ] 0
1 = 1 + 1/2 + 1/3 + 1/4 + 1/5 + …
Now Leibniz applies the line HL = (1 + y)–1 = 1 – y + y2 – y3 + ... to the line CF (= 1) to obtain
the finite area CFGLB. That is,
CFGLB = ∫01(1 + y)–1 dy = [y – y2/2 + y3/3 - y4/4... ] 0
1 = 1 - 1/2 + 1/3 – 1/4 + 1/5 + …
But if we now subtract the finite space CFGLB from the infinite space ACBEM, we obtain
Leibniz’s Syncategorematic Infinitesimals R. T. W. Arthur
6
ACBEM – CFGLB = (1 - 1) + (1/2 + 1/2) + (1/3 – 1/3) + (1/4 + 1/4) + (1/5 – 1/5) + …
= 1 + 1/2 + 1/3 + 1/4 + 1/5 + …
= ACBEM
Leibniz comments:
This is pretty amazing (satis mirabilis), and shows that the sum of the series 1, 1/2, 1/3
etc. is infinite, and consequently that the area of the space ACBGM remains the same
even when the finite space CBGF is taken away from it, i.e. that nothing noticeable is
taken away.
By this argument it is concluded that the infinite is not a whole, but only a fiction; for
otherwise the part would be equal to the whole. (A VII 3, N. 3810, p.468; October
1674)
Of interest here is the close connection between the visualizable infinite whole —the
area under the hyperbola— and its expansion as an infinite series. For the infinite series is
treated as if it were a whole, as if it has a definite sum, and so forth. But in this case the
result leads to a contradiction, given the applicability of the part-whole axiom to the infinite.
Even though this establishes the fictional nature of such infinite wholes, however, this does
not mean that one cannot calculate with them; only, the viability of the resulting calculation
is contingent on the provision of a demonstration.
As an example, we can look at another seminal calculation Leibniz made involving the
hyperbolic series, in the process of answering a challenge set him by Huygens soon after his
arrival in Paris in 1672. This was to determine the sum of the series of the reciprocal
triangular numbers, 1/1, 1/3, 1/6, 1/10, 1/15, ... Leibniz achieved this by noting that the
successive terms are twice those of the differences between successive terms of the
Leibniz’s Syncategorematic Infinitesimals R. T. W. Arthur
7
hyperbolic series: 1 = 2(1 - 1/2 ); 1/3 = 2(1/2 - 1/3); 1/6= 2(1/3 - 1/4); and so forth. But the sum of
a series of such difference terms will, because of the cancellation of terms, equal the
difference between the first term of the hyperbolic series and the last. Leibniz immediately
realized the generality of this result, which he enshrined in what I have elsewhere (Arthur
2006) called his Difference Principle: “the sum of the differences is the difference between
the first term and the last” (A VII, 3, 95). As is well known, after generalization to infinite
series, this result will lead Leibniz to his formulation of the Fundamental Theorem of the
Calculus: the sum (integral) of the differentials equals the difference of the sums (the
definite integral evaluated between first and last terms), ∫ Bdx = [A]if. Thus the sum of the
first n terms of the original series,
∑i=1n Ti = 1/1 + 1/3 + 1/6 + 1/10 + 1/15 + ... + 2/(n2 + n)
will be twice the difference between the (n + 1)th and the first term of the hyperbolic series:
∑i=1n+ 1 Hi = 1/1 + 1/2 + 1/3 + 1/4 + 1/5 + ... + 1/(n + 1)
That is,
∑i=1n Ti = 2(H1 – Hn+1) = 2[1 - 1/(n + 1)]
Applying similar reasoning to infinite series, Leibniz calculates that if the sum of the
infinite series S(H) = ∑i=1n Hi = 1/1 + 1/2 + 1/3 + 1/4 + 1/5 + ..., and half the sum of the reciprocal
triangular numbers 1/2S(T) = 1/2∑i=1∞ Ti = 1/2 + 1/6 + 1/12 + 1/20 + 1/30 …, then
S(H) - 1/2S(T) = 1/2 + 1/3 + 1/4 + 1/5 + ... = S(H) – 1
giving
S(T) = 2.
Now one might of course object to this reasoning that, since the hyperbolic series is a
diverging infinite series, Leibniz has effectively cancelled infinities in subtracting its sum S(H)
Leibniz’s Syncategorematic Infinitesimals R. T. W. Arthur
8
= ∑i=1n Hi from both sides of the equation above. Clearly this would not be news to Leibniz,
given the argument for the fictionality of the infinite treated as a whole given above. And
indeed, Leibniz is sensitive to the need for rigour here. He acknowledges that this application
of the Difference Principle “ought to be demonstrated to come out in the infinite” (362), and
proceeds to show how this can be done. This is crucial, for in the above reasoning Leibniz has
treated the infinite series as if they are wholes. But this can only be done, according to his
commentary on Galileo, if there is a corresponding demonstration. Moreover, treating the
infinite series as wholes is equivalent to treating them as if they had last terms, since
otherwise the Difference Principle would not be applicable. This gives the all-important
connection between Leibniz’s doctrines of the fictionality of infinite wholes and the
fictionality of infinitesimals. A justification of treating the infinite series as wholes is at the
same time a justification of treating them as if they had a last, infinitely small term or
terminatio, 1/∞.
Leibniz’s demonstration proceeds by taking an arbitrarily small yth term as the terminatio
of the series, where “y signifies any number whatever”. For the hyperbolic series the
terminatio of a finite series of terms H(y) will be 1/y, and for the series of reciprocal
triangular numbers ∑T(y), it will be 2/(y2 + y), since the yth triangular number is (y2 + y)/2.
Thus when half the series ∑T(y) is subtracted from ∑H(y), the terminatio of the resulting
series will be 1/y − 1/(y2 + y), or (y2 + y − y)/ (y3 + y2), or 1/(y + 1). But this is the terminatio
for ∑H(y + 1). Leibniz does not complete this demonstration in these terms, preferring to
proceed to a geometrical depiction in terms of a triangle, but it entails that for arbitrarily
large y, the sum of half the series, 1/2∑T(y), is 1 − 1/(y + 1). So the sum of the reciprocal
triangular numbers, ∑T(y), approaches 2 arbitrarily closely as y is taken arbitrarily large.
Correspondingly, the terminatio of either this series or the hyperbolic series is not actually 0,
but an arbitrarily small number.
Leibniz’s Syncategorematic Infinitesimals R. T. W. Arthur
9
This, then, yields the link between Leibniz’s syncategorematic interpretation of the
infinite and his interpretation of infinitesimals as fictions. To treat the infinite series as whole
is to treat it as if it has an infinitieth term or infinitely small terminatio; whereas in fact the
number of terms is greater than any number that can be given, and the magnitude of the
terminatio is correspondingly smaller than any that can be given. This connection is stated
explicitly by Leibniz in an annotation he made on Spinoza’s Letter on the Infinite in the Spring
of 1676:
Finally those things are infinite in the lowest degree whose magnitude is greater than
we can expound by an assignable ratio to sensible things, even though there exists
something greater than these things. In just this way, there is the infinite space
comprised between Apollonius' Hyperbola and its asymptote, which is one of the most
moderate of infinities, to which there somehow corresponds in numbers the sum of
this space: 1/1 + 1/2 + 1/3 + 1/4 + ..., which is 1/0. Only let us understand this 0, or
nought, or rather instead a quantity infinitely or unassignably small, to be greater or
smaller according as we have assumed the last denominator of this infinite series of
fractions, which is itself also infinite, smaller or greater. For a maximum does not
apply in the case of numbers. (A VI 3, 282; LoC 114-115)
For some time Leibniz appears to have hesitated over this interpretation, and as late as
February 1676 he was still deliberating about whether the success of the hypothesis of
infinities and the infinitely small in geometry spoke to their existence in physical reality too.
But by April of the same year the syncategorematic interpretation is firmly in place, and we
find Leibniz exploring the implications of the rejection of the infinitely small in a series of
mathematico-physical reflections.3 During the same period he was putting the finishing
touches to his treatise on the infinitesimal calculus, De quadratura arithmetica (Leibniz,
3 See Arthur (2009) for a discussion of the development of Leibniz’s early views on infinitesimals.
Leibniz’s Syncategorematic Infinitesimals R. T. W. Arthur
10
1993), which he submitted for the Académie Française in the summer of 1676. Here we find
his first explicit exposition of the interpretation of infinites and infinitesimals as fictions, and
the provision of a theorem which, in Leibniz’s words, “serves to lay the foundations of the
whole method of indivisibles in the soundest way possible”. Indeed, as Eberhard Knobloch has
remarked, this theorem is a “model of mathematical rigour” (2002, 72).
In the treatise Leibniz promotes his new method of performing quadratures directly
“without a reductio ad absurdum” (Prop 7, Scholium; Leibniz 1993, 35), by what we would
now call a direct integral. This, he believes, necessarily involves the assumption of “fictitious
quantities, namely the infinite and the infinitely small” (35). The traditional Archimedean
method of demonstration was by a double reductio ad absurdum: it would be shown that a
contradiction could be derived on the assumption that the quantity in question was smaller
than a given value, and another contradiction on the assumption that the quantity in question
was greater than that value, thus proving that it equalled it. Leibniz’s method is instead to
proceed by an application of the Archimedean Axiom. This axiom (actually due to Eudoxus)
asserts that for any two geometric quantities x and y (with y > x), a natural number n can be
found such that nx > y. This also entails that no matter how small a geometric quantity is
given, a smaller can be found, and it is this property to which Leibniz appeals. Thus he
prefers a justification “which simply shows that the difference between two quantities is
nothing, so that they are then equal (whereas it is otherwise usually proved by a double
reductio that one is neither greater nor smaller than the other)” (35). That is, he applies the
Archimedean axiom to demonstrate that the error involved in calculations with infinitely
small differences can be reduced to a quantity less than any given quantity by taking a
difference sufficiently small, rendering it effectively null.
Moreover, this justification does not have to be effected in every case: it is enough to
show that it can be done in a general case. This Leibniz does in a case that is surprisingly
Leibniz’s Syncategorematic Infinitesimals R. T. W. Arthur
11
general, given the usual accusations about the parlous lack of justification he and Newton are
alleged to have provided for their methods. For the key theorem that Leibniz successfully
demonstrates in De quadratura arithmetica using this Archimedean method is what is now
known as Riemannian Integration, as Eberhard Knobloch has shown in detail (2002).4 This is
Proposition 6; (Leibniz provides a similar justification for his Theorems 7 and 8). In his
demonstration of Proposition 6 (1993, 30-33), Leibniz first identifies and then relates the sum
of the “elementary rectangles [rectangula elementales]” (that is the Riemannian sum of
unequal rectangles by which the curve is being approximated, which we may denote Q), and
that of the mixtilinear figure [spatium gradiformis] that is the area under the curve between
two ordinates 1L and 4D, which we may denote A. Then he demonstrates that the difference
between the area and the sum of the elementary rectangles, A – Q, can be no greater than
the area of a certain rectangle whose height is the maximum height ψ4D of any of the
elementary rectangles, and whose width is the distance between the two ordinates 1L and 4D.
Thus A – Q ≤ 1L4D × ψ4D. But because the curve is assumed continuous, Archimedes’ Axiom
applies. Thus the height ψ4D, even though it is greater than the heights of all the other
elementary rectangles, “can be assumed smaller than any assigned quantity, for however
small it is assumed to be, still smaller heights could be taken.” Therefore the area of the
rectangle 1L4Dψ “can also be made smaller than any given surface”. Thus the difference A – Q
too “can also be made smaller than any given quantity. QED.” (30-33) There is therefore no
error involved in calculating the quadrature as the sum of an infinity of infinitesimal areas,
provided this is understood to mean that there are more little finite areas than can be
assigned, and that their magnitude is smaller than any that can be assigned; in that case, so
4 The exposition I give is indebted to Knobloch’s (2002), but in the symbolization I have followed Leibniz’s own from the De quadratura (Leibniz 1993).
Leibniz’s Syncategorematic Infinitesimals R. T. W. Arthur
12
many such areas of sufficient smallness can be taken as to make the difference between their
sum and the quadrature less than any that can be assigned.
Thus it is not the case that Leibniz has two methods, one committed to the existence of
infinitesimals as “genuine mathematical entities”, “fixed, but infinitely small”, and the other
Archimedean, treating the infinitely small as arbitrarily small finite lines, ones that could be
made as small as desired;5 nor that he simply uses the infinitesimal calculus and then airily
refers to the fact that one could instead have used an Archimedean method.6 Rather,
Leibniz’s use of the Archimedean Axiom justifies proceeding as if there are infinitesimals, and
at the same time demonstrates that what they really stand for are finite quantities which can
be taken as small as desired. Once this is demonstrated in a suitably general case, it also
justifies the use of these fictions in other analogous cases. As Leibniz himself writes, “Nor is it
necessary always to use inscribed or circumscribed figures, and to infer by reductio ad
absurdum, and to show that the error is smaller than any assignable; although what we have
said in Props. 6, 7 & 8 establishes that it can easily be done by those means.” (Scholium to
Prop. 23, Leibniz 1993, 69).
In effect, the application of the Archimedean Axiom enables a kind of Arithmetic of the
Infinite. In his article, Knobloch identifies a number of rules which are tacitly applied by
Leibniz in De quadratura arithmetica, “without demonstrating them, only relying on the ‘law
of continuity’” (2002, 67). Examples are “1. Finite + infinite = infinite”, “2.1 Finite ±
infinitely small = finite”, “2.2 x = (y + infinitely small) ⇒ x – y ≈ 0 (is unassignable)”. 2.2 can
5 Thus Henk Bos, in his classic article on Leibniz’s differentials (1974-75, 55), sees Leibniz as pursuing “two different approaches to the foundations of the calculus; one connected with the classical methods of proof by ‘exhaustion’, the other in connection with a law of continuity.” The quotations given in my text are from Bos 1974-75, 12, 13. See also Herbert Breger’s similar criticisms of Bos on these points in his (2008, 195-197). 6 Thus, to give a recent example of this kind of interpretation, Douglas Jesseph writes: “Leibniz often made grand programmatic statements to the effect that derivations which presuppose infinitesimals can always be re-cast as exhaustion proofs in the style of Archimedes. But he never, so far as I know, attempted anything like a general proof of the eliminability of the infinitesimal …” (Jesseph 2008, 233).
Leibniz’s Syncategorematic Infinitesimals R. T. W. Arthur
13
be demonstrated by Leibniz’s method as follows. Suppose x = y + dx, where dx > 0, and
suppose dx is a smallest assignable difference between x and y. Assuming it is a variable
geometric quantity, the Archimedean Axiom will apply to it. It will therefore have an
assignable ratio to another geometric quantity z such that a number n > 0 can be chosen to
make ndx > z. Therefore z/n will be smaller than dx, contrary to supposition. Therefore x – y
= dx is unassignable; the difference between x and y is smaller than any quantity that can be
assigned; it is incomparable with (has no finite ratio to) any finite quantity. What this means
is that there can be no assignable error in equating x – y with 0. Similar demonstrations can
be given for the other rules.
Leibniz was quite explicit about this Archimedean foundation for his differentials as
“incomparables”. In his response to Bernard Nieuwentijt’s criticisms in 1695 he wrote that
when dx is added to x, the increment dx
… cannot be exhibited by any construction. That is, I regard only those homogeneous
quantities to be comparable (in agreement with Euclid, Book V, Definition 5), of which
one can exceed the other when multiplied by a number, that is, a finite number. And I
maintain that any entities whose difference is not such a quantity are equal. (…) This
is precisely what is meant by saying that the difference is smaller than any given.
(Leibniz 1695; GM V 322; also quoted in Bos 1974-75, 14)
The full “Law of Continuity” that Leibniz published for the first time in 1688 is a
generalization of the method that we have just been examining:
When the difference between two instances in a given series, or in whatever is
presupposed, can be diminished until it becomes smaller than any given quantity
whatever, the corresponding difference in what is sought, or what results, must of
Leibniz’s Syncategorematic Infinitesimals R. T. W. Arthur
14
necessity also be diminished or become less than any given quantity whatever. (A VI
4, 371, 2032)7
But in a second formulation it also appears to be a generalization of the approach to
infinite series we examined above, where the infinite limiting term or terminatio is included
as if it were an infinitieth term in the series:
In any proposed continuous transition that ends in a certain limiting case (terminus), it
is permissible to formulate a general reasoning in which that final limiting case is
included.8
It is the Law of Continuity in this form that is the basis of Leibniz’s attempts to ground the
rules of the calculus. His first attempt, in his first publication of the calculus, Nova Methodus
pro Maximis et Minimis (1684)9, was not clear to his contemporaries, as it was vitiated by a
number of errors. In this article Leibniz rather confusingly begins by defining dx as “any
straight line assumed arbitrarily”, before giving his rules for manipulating of differentials
(e.g. d(xv) = xdv + vdx).10 But in a reply to the criticisms of Nieuwentijt and others, Cum
prodiisset, drafted in or shortly after 1701 but first published by Gerhardt in 1846, he
attempts “to show a little more clearly that the (so-called) algorithm of our differential
calculus proposed by me in 1684 may be confirmed to be perfectly correct”.11
7 Leibniz adds: “Or, to put it more commonly, when the cases (or given quantities) continually approach one another, so that one finally passes over into the other, the consequences or events (or what is sought) must do so too.” (A VI 4, 371, 2032; translations mine.) 8 Leibniz 1846, 40; Bos 1974-75, 56. 9 Nova methodus pro maximis et minimis, itemque tangentibus, quae nec fractas nec irrationales quantitates moratur, et singulare pro illis calculi genus, Acta Eruditorum Lips. 1684, GM V 220-226 10 Nova Methodus, GM V 220. 11 Henk Bos has drawn attention to the importance of Cum prodiisset in his article on Leibniz’s differentials (Bos 1974-75). As he observes, since it contains a reference to a work by Gouye of 1701, and “deals with the problems which were discussed in 1701-1702, it is probable that it originated in or not much later than 1701” (1974-75, 56, n. 92). In a marginal note Leibniz mentions “the Parisians” and urges that “All this must be reviewed very carefully so that it can be published, leaving out the sourer things said in contradicting others.” (Leibniz 1846, 39)
Leibniz’s Syncategorematic Infinitesimals R. T. W. Arthur
15
A
T
D
In this later justification Leibniz has dx stand for the difference between two abscissas X1
and X2, so that dx = X2 – X1, with dy the difference between the corresponding ordinates X1Y1
and X2Y2. For the sake of definiteness, he uses a particular example, the parabola x2 = ay
(Figure 2 below).12
Figure 2: Tangent to parabola
(d)x
X1 Y1
X2 Y2
Since (x + dx, y + dy) is also a point on the parabola, “we will have
y + dy = (xx + 2xdx + dxdx)/a (2.1)
and subtracting y from one side and x2/a from the other, there will remain
dy/dx = (2x + dx)/a (2.2)
which is the general rule expressing the ratio of the difference in ordinates to the difference
in abscissas.” Now Leibniz appeals to his Law of Continuity:
Now, since by our postulate it is permissible to include in a general reasoning also the
case where the ordinate X2Y1, having been moved closer and closer to the fixed
ordinate X1Y1 until it finally coincides with it, it is clear that in this case dx will be
12 Leibniz 1846, 44. Note that Leibniz characteristically draws his diagrams with the abscissas and x-axis vertical and the ordinates and y-axis horizontal. I have altered his 1X, 2Y, etc. to X1, Y2, etc., for ease of reading, and also corrected his dx to (d)x, as he seems to have intended, but otherwise this is his figure.
Leibniz’s Syncategorematic Infinitesimals R. T. W. Arthur
16
equal to zero or should be omitted, and so it is clear that, since in this case TY1 is the
tangent, X1Y1 to TX1 is as 2x to a. (Leibniz 1846, 44)
Leibniz’s summary of how the Law of Continuity is applied here is notably clear:
Hence it may be seen that in all our differential calculus there is no need to call equal
those things that have an infinitely small difference,13 but those things are taken as
equal that have no difference at all, provided that the calculation is supposed to have
been rendered general, applying equally to the case where the difference is something
and to where it is zero; and only when the calculation has been purged as far as
possible through legitimate omissions and ratios of non-vanishing quantities until at
last application is made to the ultimate case, is the difference assumed to be zero.
(44-45)
It can be appreciated, I think, how close this is to a modern justification of differentiation in
terms of limits. Leibniz has here assumed that the difference dx is a finite difference (we
would write Δx), and when an expression for Δy/Δx has been found in which Δx has as far as
possible been cancelled, he effectively takes the limit as Δx → 0 to arrive at an expression
for dy/dx.
The rigour of Leibniz’s algorithm can be seen by applying it to the counterexample
proposed by Rolle. He supposed a parabola y2 = ax, and quoted Leibniz’s result that
2y dy = a dx or (adx – 2ydy) = 0 (2.3)
Now Rolle reasoned that since (x + dx, y + dy) is also a point on the parabola, we will have
(y + dy)2 = ax + adx = y2 + 2ydy + dy2 (2.4)
⇒ dy2 = (ax – y2) + (adx – 2ydy) = 0 + 0 (2.5)
13 This is a response to Nieuwentijt’s definition of an infinitesimal as something actually infinitely small: see below.
Leibniz’s Syncategorematic Infinitesimals R. T. W. Arthur
17
Thus, he concluded, dy = dx = 0, and Leibniz’s infinitesimals are identically equal to zero.
But the use of (2.3) in the calculation violates Leibniz’s proviso that the difference may
be assumed to be zero only after “the calculation has been purged as far as possible through
legitimate omissions and ratios of non-vanishing quantities” and one has reached the limiting
case. Prior to that, the general rule that applies to all finite differences is dx/dy = (2y +
dy)/a; and if one substitutes that formula into equation (2.4) in Rolle’s calculation one
obtains the identity dy2 = dy2. If, on the other hand, having obtained dx/dy = (2y + dy)/a, one
applies it to the ultimate case by effectively letting dy → 0, one obtains (2.3) as a result.
What this application of the Law of Continuity legitimates, then, is proceeding as if dx
and dy are infinitely small quantities that can be neglected in the last step of the calculation;
whereas what in fact they stand for are finite differences that are assumed to vary in
accordance with the Archimedean Axiom, so that they can be made arbitrarily small. Leibniz
was aware, however, that this new algorithm was a kind of shorthand for what in his time
would be acceptable as a properly geometrical proof. That would have to proceed by
proportions, where only quantities of the same dimensions could stand in ratios to one
another, and all quantities would be finite or “assignable”, even in the limit. In Cum
prodiisset he therefore gave a proof sketch in which “dx and dy are retained in the
calculation in such a way that they signify non-vanishing quantities even in the ultimate
case”. He lets (d)x be “any assignable straight line” (i.e. a fixed finite line segment), and
then defines another segment (d)y, by the proportion
(d)y:(d)x = DY2:DY1 = dy:dx (2.1)
Thus (d)y varies in such a way that for all finite dx and dy > 0 it maintains the same finite
ratio to the fixed line (d)x as dy has to dx. But, as Bos observes, the same (d)y can also be
given an interpretation in the limit when the variable dx = 0, namely through the proportion
Leibniz’s Syncategorematic Infinitesimals R. T. W. Arthur
18
(d)y:(d)x = y:σ (2.2)
where σ is the subtangent to the curve. Bos’s diagram follows:
Figure 3: Leibniz’s finite surrogates
dy
(d)x
y
x dx
σ
Now, since the resulting formula is still interpretable even in the case where dx = 0, the
Law of Continuity asserts that this limiting case may also be included in the general
reasoning:14 dy:dx can be substituted for (d)y:(d)x in the resulting formulas even for the case
where dx = 0: “in this way dy and dx will always be assignable in the ratio DY2 to DY1, which
latter vanish in the ultimate case.” As Bos explains, this explication does indeed make
intelligible Leibniz’s having begun his 1684 article by having dx and dy stand for finite lines.
What was confusing is that there he had dx stand for an arbitrary finite line, corresponding to
the (d)x of this later explication.
Proceeding on this foundation Leibniz offers justifications in Cum prodiisset for the rules
of the calculus for first-order differentials that he had given in 1684: the rules for
differentiating sums and differences, products, quotients, and powers. First he constructs a
figure in which (as above) (d)x is a fixed straight line segment which remains constant as X2
14 Here I disagree with Bos, who claims that in taking the secant as the limiting position of the tangent, Leibniz “did not invoke the law of continuity; as will be seen, he used the law later, presupposing that the limiting position of the secant is the tangent.” (Bos 1974-75, 57). But that is how the law of continuity works: it does presuppose a limiting case, it does not establish its existence.
Leibniz’s Syncategorematic Infinitesimals R. T. W. Arthur
19
approaches the fixed point X1, dx = X2 – X1, dy = X2Y2 – X1Y1, and (d)y = (dy/dx)·(d)x. He also
introduces a third variable v and defines another variable straight line segment (d)v by (d)v =
(dv/dx)·(d)x. (Leibniz 1846, 46)
Following Bos, I will give as an example his justification on this basis for the
differentiation of a product: d(xv) = xdv + vdx. First Leibniz lets ay = xv (here the purpose of
the constant a is to preserve the homogeneity of the equation), and then allows x, y, and v to
increase by dx, dy, and dv respectively:
“Demonstration: ay + dy = (x + dx)(v + dv)
= xv + xdv + vdx + dxdv,
and, subtracting from each side the equals ay and xv, we have
ady = xdv + vdx + dxdv
or ady/dx = xdv/dx + v + dv
and transposing the case, as we may, to lines that never vanish, we have
a(d)y/(d)x = x(d)v/(d)x + v + dv
so that the only remaining term which can vanish is dv, and since in the case of
vanishing differences dv = 0, we have
a(d)y = x(d)v + v(d)x
as was asserted, *or (d)y:(d)x = (x + v):a.* Whence also, because (d)y:(d)x always =
dy:dx one may feign this in the case of vanishing dy, dx and put
ady = xdv + vdx.” 15
Thus d(xv) = xdv + vdx QED. (2.3)
15 Leibniz 1846, 46-47; quoted from Bos (1974-75, 58). The asterisked phrase is an error for “or (d)y:(d)x = [x(d)v/(d)x + v]:a, but this does not figure in the remaining calculation.
Leibniz’s Syncategorematic Infinitesimals R. T. W. Arthur
20
This justification may be regarded as one in keeping with the standards of rigour
expected by Huygens and Newton, and thus above reproach. It still depends on the same
conceptions of infinitesimals as fictions: they stand for quantities that are finite but variable,
and that may be made arbitrarily small.
My interpretation here contrasts with that of Bos, who reads Leibniz as having two
different interpretations of differentials, one in which they are infinitely small elements of
lines, and the other where they stand for finite lines. He points out that differentials are
equated with finite lines by Leibniz already in his first publication, the Nova Methodus of
1684, apparently in preference to the approach taken in the manuscript Elementa “which
Gerhardt identified as an alternative draft for the first publication of the rules of the
calculus, in which the differentials are introduced as infinitesimals” (Bos 1974-75, 63). But in
the Elementa, dx and dy are defined as differences of successive abscissa and ordinates which
are then assumed to become infinitely small, so that even here they are understood as limit-
elements, like Newton’s nascent and evanescent quantities (Leibniz 1848, 32-33). In the Nova
Methodus, on the other hand, Leibniz makes more explicit the finiteness of the differences
prior to this limiting process. He lets dx stand for an arbitrary finite line, and dy for a variable
finite line, whose ratio to dx in the limit as dx becomes evanescent equals the ratio of the
ordinate to the subtangent:
dy:dx = y:σ
In the more sophisticated approach of Cum prodiisset, as we saw above, (d)x is now taken as
the arbitrary finite line, and (d)y the variable finite line defined by (2.1) and (2.2), while dy
and dx are both variable finite quantities whose ratio is always equal to the ratio (d)y:(d)x,
even in the case where the finite difference dx goes to zero. Bos interprets this as meaning
that in Cum prodiisset Leibniz has defined (d)y as the differential. For if y = f(x), and the
finite differences dy and dx are written as Δy and Δx, then (2.1) becomes
Leibniz’s Syncategorematic Infinitesimals R. T. W. Arthur
21
(d)y = (Δy/Δx)(d)x = [f(x + Δx) – f(x)]/Δx·(d)x (2.4)
In the limit as Δx → 0, the expression in braces is f´(x), the derivative with respect to x of
the function f(x), and (d)y is the differential of y as defined by Cauchy (Bos 1974-75, 59).
But this, it seems to me, is to misinterpret Leibniz’s justification, the aim of which is not
to define (d)y as the differential, but rather to show that dy and dv may still have an
assignable ratio to dx even in the limit where dx → 0. In that limit, dy, dv and dx are
infinitely small, fictional quantities. These are the differentials. The role of (d)y, (d)v and
(d)x is to be “finite surrogates”: they remain assignable quantities throughout the limiting
process as dx becomes arbitrarily small, and since (d)y and (d)x have an assignable ratio equal
to dy:dx even in that limit (likewise for (d)v and (d)x), differential equations involving terms
in dy, dv and dx to the same order are interpretable under the fiction where these quantities
stand for infinitely small quantities.
Having shown this in an arbitrary case, it obviates the need to show it in every case. The
strict proof operating only with assignable quantities justifies proceeding by simply appealing
to the fact that dv is incomparable with respect to v: in keeping with the Archimedean axiom,
it can be made so small as to render any error in neglecting it smaller than any given. Thus, in
a letter to Wallis in 1699, Leibniz justifies the rule for d(xy) = xdy + ydx as follows
… there remains xdy + ydx + dx·dy. But this dx·dy should be rejected, as it is
incomparably smaller than xdy + ydx, and this becomes d(xy) = xdy + ydx, inasmuch
as, if someone wished to translate the calculation into the style of Archimedes, it is
evident that, when the thing is done using assignable qualities, the error that could
accrue from this would always be smaller that any given.16
16 “… restat xdy + ydx + dxdy. Sed hic dxdy rejiciendum, ut ipsis xdy + ydx incomparabiliter minus, et fit d, xy = xdy + ydx, ita ut semper manifestum sit, re in ipsis assignabilibus peracta, errorem, qui
Leibniz’s Syncategorematic Infinitesimals R. T. W. Arthur
22
As a second example of Leibniz’s approach to infinitesimals, it will be instructive to
examine Leibniz’s criticisms of Newton’s proof of his Lemma 9 in the Principia and the
demonstration he offers in its stead. Here I will follow closely the development of Leibniz’s
reactions to Newton’s Principia and its methods given by Domenico Bertoloni Meli (1993), who
has analysed the evolution of Leibniz’s mathematical cosmology from his initial hostile
reactions to Newton’s to the development of his own rival dynamics of celestial motion, in a
brilliant and careful study of the surviving documents.
As Leibniz correctly recognized, Lemmas 9-11 are crucial for Newton’s proofs of the
inverse square law; Lemma 10 in particular, is a corollary of Lemma 9, and is appealed to by
Newton (in the first edition of the Principia) in the proof of Proposition 6. Each of Newton’s
lemmas is an instance of his Method of First and Last Ratios, and establishes a ratio between
quantities as they are on the very point of being generated or vanishing. The figures,
accordingly, are “ultimate” ones, depicting what Whiteside has aptly termed “limit motions”
(1967, 154) the motions a body would undergo during a “moment” if it continued with the
velocity it had at the beginning of that moment. Leibniz had no objection to this procedure;
but he did object to what he found in Proposition 6, namely the treatment of a curved
trajectory in such an ultimate moment as compounded from two motions, one a rectilinear
inertial motion along the tangent, and the other an acceleration towards the centre. This was
an objection of his of long standing, and is closely related to his syncategorematic
understanding of moments and of infinite division. In a piece written at the beginning of April
1676, he reasoned that if there were such a thing as a perfect fluid—by which he meant
matter divided all the way down into points, each individuated by a differing motion—then “a
new endeavour [would] be impressed at any moment whatever” on a body moving in the
inde metui queat, esse dato minorem, si quis calculum ad Archimedis stylum traducere velit.” (Leibniz to Wallis, 30th March 1699; GM IV 63).
Leibniz’s Syncategorematic Infinitesimals R. T. W. Arthur
23
fluid. But this would be to compose a curved line from points and time from instants, and
would also entail an “impossible” composition at every single instant:
But if this is conceded, time will actually be divided into instants, which is not
possible. So there will be no uniformly accelerated motion anywhere, and so the
parabola will not be describable in this way. And so it is quite credible that circles and
parabolas and other things of that kind are all fictitious.... For supposing a point
moves in a parabolic line, it will certainly be true of it that at any instant it is moving
with a uniform motion in one direction, and with a uniformly accelerated motion in
another, which is impossible. (A VI 3, 492; LoC 74-77)17
When he first confronted Newton’s Principia some 12 years later, Leibniz’s initial reaction
was therefore to believe that there was a mistake in Newton’s composition of motions in
deriving Proposition 6. On his understanding, in an ultimate moment only straight line motions
(and equivalently the lines they would traverse in such a moment) can be compounded, since
ultimately the curve is resolved into an infinite-sided polygon with fictional straight sides,
each one representing a geometric “indivisible” or ultimate difference between successive
values. Seeing the dependence of Proposition 6 on Lemmas 9 and 10, therefore, he
concentrated his attention on them.
In Lemma 9, Newton has a curved trajectory depicted in an ultimate moment, and a
claim is made that “the areas of the [curvilinear] triangles ADB and AEC will ultimately be to
each other in the squared ratio of the sides”, i.e. ∆ADB:∆AEC = AD2:AE2 (Newton 1999, 437).
Newton performs his proof by using his Method Of Finite Surrogates: here these surrogates are
17 Here Leibniz appears to be applying a principle he had just formulated: “If a given motion can be resolved into two motions, one of them possible and the other impossible, the given motion will be impossible.” (A VI 3, 492; LoC 72-73) The quoted passage should be compared with the quotation Bertoloni Meli gives from Leibniz’s letter to Claude Perrault, also written in 1676: “I take it as certain that everything moving along a curved line endeavours to escape along the tangent of this curve; the true cause of this is that curves are polygons with an infinite number of sides, and these sides are portions of the tangents…” (quoted from Bertoloni Meli 1993, 75).
Leibniz’s Syncategorematic Infinitesimals R. T. W. Arthur
24
the lines Ae and Ad, which are finite, with Ae remaining fixed, and in proportion to AD and
AE, which are supposed variable; likewise Abc is a fixed portion of curve similar to ABC, with
the elements of the curve AB and AC supposed variable, gradually shrinking toward A until B
and C “come together” with it. The line AGg is tangent to both curves at A. Newton’s proof
then proceeds by noting that in the limit A, B and C coincide, ∠cAg vanishes, and “the
curvilinear areas Abd and Ace will coincide with the rectilinear areas Afd and Age, and thus
(by Lemma 5) will be in the squared ratio of these sides Ad and Ae” (437). Therefore, given
the assumed proportionality between these finite surrogates and the infinitesimals, “the
areas ABD and ACE also are ultimately in the squared ratio of the sides AD and AE. QED”
(437).
Leibniz’s Syncategorematic Infinitesimals R. T. W. Arthur
25
e g c Figure 4: Newton’s Lemma 9
d b
E C
D B
A
Leibniz’s reaction to Newton is fraught with irony. For he neglects Newton’s use of
finite surrogates, which is to all intents and purposes identical to his own way of justifying
the calculus in “Cum prodiisset” using only assignable quantities, as explained above. (This
irony is compounded by the fact that Leibniz rather arbitrarily replaces Newton’s letters E
and C by (D) and (B), the same bracket notation he usually uses for finite lines.) It is possible,
of course, that Leibniz adopts his finite surrogate technique after further reflection on
Newton’s method, although, as Bos notes, Leibniz had promoted his differentials as finite
variable quantities bearing a proportion to fixed lines in his first publication on the calculus
(1684), (albeit in a form that was not easy for his contemporaries to understand).18 In any
case, in these first notes on Newton’s Principia Leibniz dispenses with the finite surrogates
completely, so as to concentrate on the curvilinear figures that Newton has depicted in his
“ultimate moment”. Here is his diagram, statement and criticism of Newton’s Lemma 9,
following the description given by Bertoloni Meli:
18 See Bos (1974-75, 57, 63). Certainly some of the wording of Leibniz’s later justifications seems to show Newton’s influence, as Guicciardini notes in his excellent discussion (1999, 161), even if he had developed the method of finite surrogates independently.
Leibniz’s Syncategorematic Infinitesimals R. T. W. Arthur
26
(D) (B) D B Figure 5: Leibniz: Newton’s Lemma 9 A
If AB, A(B) are unassignable, ΔADB and ΔA(D)(B) will be in the squared ratio of the
sides (+ namely, it must be understood that the Δs are similar, and so angle BA(B) has
a ratio to angle DAB that is infinitely small, but this does not seem to be true, since in
ΔAB(B) the side B(B) has an assignable ratio to AB or A(B), and so to AD and DB.+)
That is, Leibniz objects that in the limit Δs ADB and A(D)(B) will not be similar, since
Newton’s diagram already represents the ultimate moment, and in it angle BA(B) has a finite
rather than infinitesimal ratio to the sides of the triangles. In an effort to show that Newton
has made an error in calculating with infinitesimals, Leibniz proceeds with his own
calculation, in which the infinitesimal order of the angle BA(B) with respect to AB and AD is
rendered explicit by setting AD = x, DB = y, A(D) = x + dx, and D(B) = y + dy. Leibniz does not
succeed in demonstrating any error on Newton’s part; nor in fact is there an error, as Meli
observes, since as x tends to zero so does dx, so that ultimately the triangles become similar.
Nevertheless, even though he fails to show anything wrong in Newton’s procedure,
Leibniz’s workings provide us with a nice illustration of his approach to the differential
calculus. First he expresses y as a series expansion in x:
y = a + bx + cx2 + ex3 + .... (2.4)
Leibniz’s Syncategorematic Infinitesimals R. T. W. Arthur
27
Now since x = 0 when y = 0, we have a = 0 —a point Leibniz first realized, and then
subsequently forgot, rendering his calculation inconclusive. But as Meli shows, the lemma
follows fairly easily from these premises. For
xy = bx2 + cx3 + ex4 + ... (2.5)
so that since (assuming Lemma 5) area[ADB] = AD·DB/2,
area[ADB] = xy/2 = 1/2(bx2 + cx3 + ex4 + ...) (2.6)
Meli remarks that “If x becomes infinitesimal, the terms in x3, x4, ... are negligible, and the
area of ADB (= xy/2) is indeed proportional to AD2 (= x2)” (242). To this one might object that
it seems unjustified to assume terms in x3, x4, ...are negligible when terms in x2 are not. In
fact, however, we can proceed more rigorously. For although Leibniz apparently thought
Newton’s finite surrogates could be dispensed with, they are completely in accord with his
way of proceeding. Thus let DB:AD = db:Ad, etc., as Newton had assumed, with db and Ad
finite, and with this ratio remaining in a finite and non-vanishing proportion to ec and Ae
(with Ae fixed) even as x → 0. Then
DB/AD = AD•DB/AD2 = b + cx + ex2 + ... (2.7)
But since this equals db:Ad, which is finite and non-vanishing as x → 0, it follows that
2 × area[ADB)]/AD2 = AD•DB/AD2 = b (const.) (2.8)
Now the same reasoning will apply for ΔA(D)(B), giving
area[ADB)]:area[A(D)(B)] = AD2:A(D)2 (2.9)
Newton’s Lemma is therefore sound even by Leibniz’s lights, as indeed he came to realize
despite not having successfully completed this calculation.19
19 As Meli reports (Bertoloni Meli 1993, 103), Leibniz’s attitude seems to have undergone a sea change on moving on from Section 1 of the Principia on First and Last Ratios, to section 2, on the
Leibniz’s Syncategorematic Infinitesimals R. T. W. Arthur
28
Alternatively, the move from equation (2.7) to (2.8) is justifiable by the consideration
that dx is incomparably small relative to any finite quantity of the same order as x, and thus
ignoring it will produce an error smaller than any that can be assigned, and thus null. Thus if
x in equation (2.7) is an infinitesimal variable, then cx stands for a finite line that can be
made smaller than any finite quantity, and thus smaller than b, for any b and c, and a fortiori
for higher order terms.
Similar considerations apply to Lemma 11, stating that the curvature of a trajectory at a
given point is the same as the curvature of the osculating circle at that point. In the first
stage of composition of his Marginalia identified by Meli, Leibniz objects to Newton’s
Corollary 7 to Proposition 4, in which the proportionality of centripetal forces to v2/r, proved
in Proposition 4 for bodies rotating uniformly in circular orbits, is extended to non-circular
orbits for limit-motions: “Since I do not yet accept the generality of Lemma XI, I also doubt
the generality of this corollary 7”. (Bertoloni Meli 1993, 107) Somewhat later, according to
Meli, he corrected himself: “On the contrary, this is true, because the considerations on the
secant of the angle made by the radius from the centre to the curve, and that of the radius of
the osculating circle to the curve, vanish on account of the similarity of the figures.”20 That
is, according to Meli’s analysis, “since Proposition 4 is stated in the form of a proportion
between the homologous elements of two similar figures, their similarity cancels out the
dependence of paracentric conatus [—the endeavour toward the centre—] on the secant of
the angle and on the osculating radius” (107).
determination of central forces. Accordingly, in the first set of Excerpts Leibniz took from the Principia the following year, “lemma 9 is transcribed without commentary, and seems to be accepted without difficulty” (242). 20 Marginalia, M 42 A: translated from the Latin quoted by Meli, 107.
Leibniz’s Syncategorematic Infinitesimals R. T. W. Arthur
29
3. SMOOTH INFINITESIMAL ANALYSIS
Smooth Infinitesimal Analysis has many features in common with Leibniz’s approach. It
begins, like Leibniz, by eschewing the composition of the continuum from an infinity of
points. In contrast to the account of the continuum established by Cantor based on Point Set
Theory, according to which a continuous line is an infinite (indeed, nondenumerably infinite)
set of points, SIA is rooted in Category Theory, and mappings rather than points are taken as
basic. The category of smooth manifolds, Man, is embedded in an enlarged category C which
contains “infinitesimal” objects, and a topos SetC is then formed of sets varying over C. “Each
smooth topos E is then identified as a certain subcategory of SetC. Any of these toposes has
the property that its objects are undergoing a form of smooth variation, and each may be
taken as a smooth world.” (Bell 1998, 14). With this foundation secured, a great many results
of differential calculus may be obtained “—with full rigour— using straightforward calculations
with infinitesimals in place of the limit concept” (Bell, 1998, 4), as if infinitesimals exist.
Here I say “as if infinitesimals exist” advisedly. For another feature that SIA has in
common with Leibniz’s approach is that, whilst it licenses certain infinitesimal techniques, it
is not committed to the existence of infinitesimals in the continuum. That is, as in Leibniz’s
theory, infinitesimals are fictions in a precisely defined sense. The sense in which they are
fictions in SIA, however, is that although it is denied that an infinitesimal neighbourhood of a
given point, such as 0, reduces to zero, it cannot be inferred from this that there exists any
point in the infinitesimal neighbourhood distinct from 0.21 Thus the Law of Excluded Middle,
and with it the Law of Double Negation, both fail in smooth worlds. Bell explains this as
follows. Define two points a and b on the real line (as represented in a smooth world Σ) as
21 In a sympathetic but exacting analysis of SIA, Geoffrey Hellman observes that the status of infinitesimals in SIA is odd. On the one hand, “the "existence of non-zero nilsquares" is actually refutable”, but on the other, “in the background one assumes their ‘possibility’ in making sense of the whole theory” (2006, 10). Certainly there is no constructive definition of them (11), even as fictional entities. Here, I think, Leibniz’s theory is more in keeping with the constructivist spirit.
Leibniz’s Syncategorematic Infinitesimals R. T. W. Arthur
30
distinguishable iff they are not identical, i.e. iff ¬a = b, where ‘=’ denotes identity. Now
define the infinitesimal neighbourhood Ι(0) of a given point 0 as the set of all those points
indistinguishable from 0. That is, define Ι(0) as follows:
Ι(0) =def x | ¬¬x = 0 (3.1)
Now if the Law of Double Negation (or, equivalently, the Law of Excluded Middle) held in
Σ, we could infer that x = 0 for each x in Ι(0), so that the infinitesimal neighbourhood of 0 Ι(0)
would reduce to 0. But we know that this neighbourhood does not reduce to 0 in Σ. So we
cannot infer the identity of points from their indistinguishability. Again, suppose there is a
point a in Ι that is distinguishable from 0, i.e. suppose there is a point a ε Ι such that ¬a = 0.
But since a ε Ι, ¬¬a = 0 by definition. But this is a contradiction. Therefore it is not the case
that there exists a point a in Ι that is distinguishable from 0. That is, the logic of smooth
worlds is intuitionistic logic: the Law of Non-contradiction holds, as we have just seen; but
the Laws of Excluded Middle, Double Negation, and one form of Quantifier Negation do not
hold in smooth worlds. From ¬(∀x ∈ Ι)x = 0 (it is not the case that all members of Ι are
identical with 0), we can not infer that (∃x ∈ Ι)¬x = 0 (there is a member of Ι that is
distinguishable from 0).22 This makes precise the older conception of an infinitesimal
difference as a difference smaller than any assignable, but not zero. It does so, in effect, by
denying that an unassignable difference reduces to zero, but not allowing the inference from
this that there exists an unassignable difference different from zero. it is in this sense that
the infinitesimal intervals of SIA are fictional.
On this foundation Bell erects the theory of SIA. First he defines Δ as consisting in those
points x in R such that x2 = 0. The letter ε then denotes a variable ranging over Δ (21). The
fundamental assumption is then that every curve is microstraight (9, 22). That is, arbitrary 22 More precisely, the logic of smooth toposes is free first-order intuitionistic or constructive logic. See Bell 1998, 101-102.
Leibniz’s Syncategorematic Infinitesimals R. T. W. Arthur
31
functions f: R → R are assumed to behave locally like polynomials, so that with f(x) = a0 + a1x
+ a2x2 + ... anxn and ε2 = 0, we have
f(ε) = a0 + a1ε for any ε in Δ. (3.2)
This is the Principle of Microstraightness, and it implies that (ε,f(ε) lies on the tangent to the
curve at the point (0,a0). It also entails (if one considers only the restriction of g of f to Δ)
that g is affine on Δ. This may be compared with Leibniz’s statement that “… to find a
tangent is to draw a straight line which joins two points of the curve which have an infinitely
small distance, that is, the prolonged side of the infinitangular polygon which for us is the
same as the curve.” (Leibniz 1684a, GM V 223). That consideration motivates the Principle of
Microaffineness (Bell 1998, 23):
For any map g: Δ → R, there exists a unique b in R such that, for all ε in Δ, we have
g(ε) = g(0) + εb. (3.3)
This principle allows one to define the derivative of an arbitrary function f: R → R as follows
(1998, 26). Define the function gx(ε) = f(x + ε). By Microaffineness it follows that there is a
unique bx such that for all ε in Δ,
f(x + ε) = gx(ε) = gx(0) + εbx = f(x) + εbx (3.4)
If we allow x to vary, the values bx will constitute a new function, the derivative f´(x):
f(x + ε) = f(x) + εf´(x) (3.5)
If the function f is a function of time, and ε is an infinitesimal time, we have as a direct
consequence of this the Principle of Microuniformity (of natural processes): any such process
may be considered as taking place at a constant rate over the “timelet” ε (1998, 9). Another
Leibniz’s Syncategorematic Infinitesimals R. T. W. Arthur
32
important consequence of Microaffineness is the Principle of Microcancellation: for any a, b
in R, if εa = εb for all ε in Δ, then a = b (1998, 24). I quote Bell’s proof:
Suppose that, for all ε in Δ, all εa = εb and consider the function g: Δ → R defined by
g(ε) = εa. The assumption then implies that g has both slope a and slope b: the
uniqueness clause in Microaffineness yields a = b. (24)
I will now illustrate the neatness and simplicity of SIA by means of two examples: (i) the
proof of the product rule for derivatives, and (ii) the proof of Newton’s Proposition 1 of the
Principia, the Kepler Area Law. For both of these illustrations I shall follow Bell’s exposition.
(i) The Product Rule for Derivatives
Let y(x) = f(x)g(x). We wish to prove that y´ = f´g + f g´. By the definition of the derivative,
f(x + ε) = f(x) + εf´(x), (3.6)
and likewise for g and y. Thus since
y(x + ε) = f(x + ε)·g(x + ε), (3.7)
we have
y(x) + εy´(x) = [f(x) + εf´(x)]·[g(x) + εg´(x)] (3.8)
= y(x) + ε[f´(x)g(x) + g´(x)f(x)] + ε2 f´(x)g´(x) (3.9)
Now subtracting y from both sides and assuming ε2 = 0, we have
εy´(x) = ε[f´(x)g(x) + g´(x)f(x)], and thus (3.10)
y´(x) = f´(x)g(x) + g´(x)f(x) (3.11)
The inference from (3.10) to (3.11) is guaranteed by the Principle of Microcancellation: if εa =
εb for all ε in Δ, then a = b. As we saw, this in turn depends on the nilsquare property ε2 = 0.
Leibniz’s Syncategorematic Infinitesimals R. T. W. Arthur
33
So in the proof, this property is invoked three times: indirectly in the definition of the
derivative (3.6), in the inference from (3.9) to (3.10) directly, and again indirectly in proving
the Microcancellation property involved in the inference from (3.10) to (3.11).
(ii) Newton's Proposition 1, the Kepler Area Law
In his proof of this law, Bell assumes that the area A, the radius r and the angle θ are all
functions of t, which increases by a nilsquare infinitesimal ε “with ε in Δ”. “Then by
Microstraightness the sector OPQ is a triangle of base r(t + ε) = r + εr´ and height
r sin[θ(t + ε) – θ(t)] = r sinεθ´ = rεθ´” (Bell 1998, 69) (3.12)
Here the nilsquare property is invoked in Microstraightness, in the definitions of the
derivatives r´ and θ´, and again in the equating of sinεθ´ with εθ´. Since the area is half the
base times the height, this gives for the area of sector OPQ
OPQ = 1/2(r + εr´)rεθ´ = 1/2εr2θ´ (3.13)
if the term 1/2ε2rr´θ´ is dropped, again invoking the nilsquare property. But the area OPQ is
the increment in area produced by the motion of the radius vector,
OPQ = A(t + ε) – A(t) = εA´(t) (3.14)
so that (3.13) and (3.14) give, by Microcancellation,
A´(t) = 1/2r2θ´ (3.15)
Expressing A´ as a function of x, x´, y, y´, Bell then proves that A´´(t) = 0, so that, assuming
A(0) = 0, we have the Area Law
A(t) = kt, (3.16)
where k is a constant.
Leibniz’s Syncategorematic Infinitesimals R. T. W. Arthur
34
As a historical note, we may remark that Bell’s procedure is reminiscent of that
pioneered by the seventeenth century Dutch mathematician Bernard Nieuwentijt.23 In his
treatment (which was independent of Newton’s and Leibniz’s) Nieuwentijt began by defining
infinitesimal and infinite quantities: “A quantity smaller than any given one may be called,
for the sake of abbreviation, infinitesimal; greater than any given one, infinite.” He then laid
down a number of axioms. The first is that “That which cannot be multiplied so many times,
that is, by such a great number, that it becomes equal in magnitude to any given [finite]
quantity, however small, is not a quantity, but a mere nothing in geometry.” (Nieuwentijt
1695, 2: Axiom 1). Axiom 2 states that any arbitrary finite quantity can be divided into
arbitrarily many equal or unequal parts less than any given quantity, so that the division of a
finite quantity b by an infinite number m yields an infinitesimal quantity. (This is in accord
with Axiom 1, since b/m may be multiplied by the product of the infinite number m and the
finite number c/b so that it does equal any other finite quantity c.) But it now follows
(Nieuwentijt 1695, 4: Lemma 10) that if two infinitesimal quantities b/m and c/m are
multiplied together, their product bc/mm is zero. For when multiplied by the largest possible
number m, the product bc/m is still infinitesimal, and therefore cannot be made equal to any
finite quantity; by Axiom 1, the product of any two infinitesimal quantities is therefore zero.
Nieuwentijt’s infinitesimals are nil-square infinitesimals. Nieuwentijt was proud of this result,
regarding it as the “Ariadne’s thread” that would provide an exit from the labyrinth of the
continuum by establishing a strict criterion for which infinitesimal quantities could indeed be
neglected in any calculation (Nieuwentijt 1995, 81).
Notwithstanding this agreement on nilsquare infinitesimals, there are, of course, many
profound differences between SIA and Nieuwentijt’s approach. For SIA, like Leibniz’s
approach, is based on smoothly varying geometric quantities, whereas Nieuwentijt’s
23 See the lucid account of Nieuwentijt’s theory in Mancosu 1996, 158ff, and also Nagel 2008, 200-205, for a succinct account of his dispute with Leibniz.
Leibniz’s Syncategorematic Infinitesimals R. T. W. Arthur
35
infinitesimals are defined through division by an infinite number m, where m is the largest
number. They are therefore invertible, unlike the infinitesimals of SIA; they are also actual
infinitesimals by Nieuwentijt’s definition, quantities smaller than any finite quantity, the
inverses of categorematically infinite quantities. Moreover, SIA presupposes intuitionistic
logic, whereas Nieuwentijt’s logic is classical. Finally, Nieuwentijt’s theory does not admit
higher order differentials, whereas, as we shall see, this is possible in SIA. But the key points
it has in common with Nieuwentijt’s approach are these: the Principle of Microaffineness,
which “entails that all curves are ‘locally straight’” (110); and the Nilsquare Property, which
guarantees that all squares and higher powers of an infinitesimal are intrinsically zero, and
not just by comparison with other quantities.24
4. PROPOSITIONS 4 AND 6 OF NEWTON’S PRINCIPIA
The comparison between Leibniz’s syncategorematic approach and that of SIA can be set in
relief by applying both to the same examples. Because of its historical importance, and also
because it gave both Leibniz and Varignon some measure of grief25 and therefore presents
itself as an excellent touchstone, I shall consider Newton’s Proposition 6 of Book 1 of the
Principia, the theorem that is the basis of his derivation in Proposition 11 of the inverse
square law of attraction due to gravity for a body orbiting in an ellipse. I will briefly present
Newton’s own proof of Proposition 6 with some commentary. Then I will proceed to a sketch
of Leibniz’s proof of the inverse square law for the ellipse, in which he avails himself of
Newton’s Proposition 1 (the area law), and its composition of motions along the chords, to
give an algebraic proof using his method of differentials. But first, to motivate this, I will
follow Nico Bertoloni Meli in giving an analysis of Newton’s Proposition 4 according to the two 24 Mancosu has very aptly summarized the difference between Nieuwentijt’s approach and Leibniz’s in his (1996, 160). One of these differences is that on Nieuwentijt’s theory, the infinitesimal b/m cannot be eliminated from computations; but Bell’s Principle of Microcancellation shows how to circumvent this problem. Bell has also provided a useful comparison of the difference between Nieuwentijt’s approach and Leibniz’s in his (2006, 98-100). 25 On Varignon and Leibniz on central force see Bertoloni Meli (1993, 81-83, 201ff.).
Leibniz’s Syncategorematic Infinitesimals R. T. W. Arthur
36
different compositions of motion: Newton’s composition of a rectilinear uniform motion along
the tangent with a uniform acceleration towards the centre, and Leibniz’s composition of a
rectilinear uniform motion along the chord with a rectilinear uniform motion towards the
centre (Bertoloni Meli 1993, 82-83).
It seems appropriate to consider Newton’s treatment in the first edition of the
Principia (published in 1687), since this is the one that Leibniz’s critical comments were
directed towards.26 There he states Proposition 6 as follows:
Figure 6: Newton’s Proposition 6 R P Q T Z S A
If a body P, revolving about a centre S, describes any curved line APQ, while the
straight line ZPR touches that curve at any point P; and to the tangent from any other
point Q of the curve, QR is drawn parallel to the distance SP, and the perpendicular
QT is dropped onto the distance SP: I say that the centripetal force is as the reciprocal
of the solid SP2•QT2/QR, provided that the quantity of this solid is always taken as that
which is made when the points P and Q come together.
In the first edition27 Newton gives the following proof:
26 For the second edition (published in 1713) Newton composed a new Proposition VI, which remains the same in the third (1726), and the first edition text became Corollary 1. Although the first edition can be pieced together from the footnotes in Newton 1999, Part I is usefully presented in its entirety (in a different translation by Mary Ann Rossi) by J. Bruce Brackenridge (1995, 235-267). 27 This proof is replaced by a new proof (of Corollary 1 to the new proposition 6) in the 2nd and 3rd editions, where QR is said to be “equal to the sagitta of an arc that is twice the length of an arc QP,
Leibniz’s Syncategorematic Infinitesimals R. T. W. Arthur
37
For in the indefinitely small figure QRPT the nascent linelet QR is, given the time, as
the centripetal force (by Law 2), and given the force, as the square of the time (by
Lem. 10), and thus, if neither be given, as the centripetal force and the square of the
time jointly, and thus the centripetal force is as the linelet QR directly and the square
of the time inversely. But the time is as the area SPQ, or its double SP × QT, that is, as
SP and QT jointly, and thus the centripetal force is as QR directly and SP2 × QT2
inversely, that is, as SP2•QT2/QR inversely. QED. (Newton 1999, 454)
In Proposition 11 Newton went on to use this formula to derive an expression for the
centripetal force holding a body in an elliptical orbit. If the minor and major axes of the
ellipse are respectively b (= 2BC) and q (=2AC), then the latus rectum of the ellipse is L =
2BC2/AC = b2/q. By adroitly using the properties of the ellipse and the lemmas he had
established for proportions in the limit motions, Newton is able to prove that SP2·QT2/QR =
L·SP2. Therefore from Proposition 6, the centripetal force is the reciprocal of this, and thus as
the inverse square of the altitude, SP.
Newton’s proof of Proposition 6 involves a composition of a uniform motion with a
uniformly accelerated motion, which Leibniz could not accept, as mentioned in section 2. He
interprets Newton’s “force” as cause of an increment in quantity of motion, so that it should
be proportional to velocity, not to acceleration.28 Newton’s proof also appeals to Lemma 10,
whose validity Leibniz initially doubted. Thus in his marginalia to Newton’s Principia Leibniz
altered Newton’s statement in the proof of proposition 6, “the nascent linelet QR is, given
the time, as the centripetal force (by Law 2), and given the force, as the square of the time
with P being in the middle; and twice the triangle SQP (or SP x QT) is proportional to the time in which twice that arc is described and therefore can stand for the time.” (453-54). Nonetheless, Newton does write after the new proof of Prop. 6: “The proposition is easily proved by lem. 10, coroll. 4.”. 28 Bertoloni Meli notes how, in his notes on Corollary 1 to the First Law, Leibniz objected to Newton’s use of “force” in stating it—writing after “… vi”, “(malo dicere conatu)”— meaning that vis should be replaced by conatus, that is (directed) infinitesimal tendency to motion. See Bertoloni Meli (1993, 224, 239, 163).
Leibniz’s Syncategorematic Infinitesimals R. T. W. Arthur
38
(by Lem. 10)”, replacing both instances of the word ‘force’ by ‘velocity’, and ‘as the square
of the time’ simply by ‘as the time’.29
Subsequently, as we noted in Section 2 above, Leibniz came to see that Newton’s way of
applying his Lemmas to limit-motions involved no mathematical error. In fact, as Meli
explains, the two ways of analyzing the composition of motion, Newton’s composition of a
rectilinear uniform motion along the tangent with a uniform acceleration towards the centre,
and Leibniz’s composition of a rectilinear uniform motion along the chord with a rectilinear
uniform motion towards the centre, generally lead to the same mathematical result. But
Leibniz still preferred a composition of two uniform motions, rejecting the physicality of
Newton’s instantaneous acceleration.30 It will be worthwhile to further our understanding of
Leibniz’s reasons for this before proceeding to his derivation of the centripetal force law for
the ellipse, by examining proofs of the centripetal force law for circular motion (Newton’s
Proposition 4) according to the two different compositions.
In his own proof of Proposition 4 in the first edition using a tangent composition, Newton
draws two homologous and concentric circles, so that the centripetal forces are “in the first
ratio of the nascent spaces CD and cd”, where CD and cd are the deflections of the two
revolving bodies. (The second homologous figure was necessary, since in the mathematics of
his day only ratios of quantities with the same dimensions could be taken.) We will dispense
with that here, and follow Meli (1993, 82-83) in giving a modernized proof of Proposition 4.
For a body orbiting S in a circular arc from P to Q we have the diagram in Figure 7 below:
29 See Meli 1993, 162. 30 “La voye est plus simple,” Leibniz wrote to Varignon in October 1706, “qui ne met pas l’acceleration dans les elemens, lorsqu’on n’en a point besoin. Je m’en suis servi depuis de 30 ans.” (GM IV 150-151; Bertoloni Meli, 1993, 81).
Leibniz’s Syncategorematic Infinitesimals R. T. W. Arthur
39
Figure 7: Tangent Composition
R P
Q
S
N
Here the centripetal force is as the deflection QR. From the figure, with the diameter of the
circle NP = 2SP, ∠NQP is a right angle, so that QR:PQ = PQ:NP = PQ:2SP, giving
QR = PQ2/2SP (4.1)
Likewise, the first ratios of the velocities along the arcs of the two circles in equal times will
be as the first ratio of the nascent arc PQ to its homologue, and therefore in the ratio of the
nascent straight lines PQ and its homologue. Again dispensing with proportions, in a given
moment dt, with v the velocity along PQ, PQ = vdt and SP = r, so the centripetal force F is as
QR = PQ2/2SP = 1/2(v2/r) dt2 (4.2)
In fact, however, the motion along QR is uniformly accelerated so that QR = 1/2a dt2. Thus
a = v2/r (4.3)
Leibniz objected to this, as he did not think there could be an acceleration in an instant.
Rather, he conceived acceleration to occur by successive increments of motion, in accordance
with the belief, typical of the mechanical philosophy, that all action is by contact. Thus while
he had no objection to the composition of motions in Newton’s Proposition 1, where the
action of a force is represented by an increment of motion occurring in a moment, and the
parallelogram law is applied to combine two uniform motions into a third, he could not
Leibniz’s Syncategorematic Infinitesimals R. T. W. Arthur
40
accept a composition of a uniform inertial motion along the tangent PR with a uniformly
accelerated motion toward the centre, along QR.
So he preferred the kind of composition used in Proposition 1, where the curved
trajectory is conceived as the “ultimate perimeter” of a polygon as the number of sides is
increased indefinitely and their width is decreased indefinitely. As is clear from the figure
below, with QR and SP parallel and equal, QR/PQ = PQ/SP, so that
QR = PQ2/SP. (4.4)
Figure 8: Chord Composition
R P O
Q S
But SP is finite and PQ is first order infinitesimal; therefore QR must be second-order
infinitesimal. Thus the velocity w of the body along QR, although constant because it
represents a uniform motion, must be a first order differential du, so that
QR = w•dt = du•dt = d(dr/dt)•dt = ddr (4.5)
Meanwhile
PQ2/SP = v2dt2/r (4.6)
Combining (4.4), (4.5) and (4.6) we have
ddr = v2dt2/r (4.7)
On a modern understanding, we would interpret this as a formula for acceleration:
Leibniz’s Syncategorematic Infinitesimals R. T. W. Arthur
41
d2r/dt2 = v2/r (4.8)
This is not, however, how Leibniz would understand (4.7). For him it says that the solicitation
in equal infinitesimal times dt is as v2 and inversely as r. Moreover, as Leibniz explains in the
Tentamen de Motuum Coelestium Causis published in the Acta of Leipzig in February 1689, “if
motion is represented by a finite line which is traversed by a body in a given time, the
impetus or velocity is expressed by an infinitely small line, and the element of velocity, which
is solicitation of gravity or centrifugal conatus, by a line infinitely many times infinitely
small”.31 That is to say, on Leibniz’s way of thinking, a body only has motion in a finite time.
The instantaneous velocity v is represented as a line-element dr produced in an element of
time dt whose ratio dr:dt is equal to the ratio of two finite lines described in the same finite
time. Solicitation, on the other hand, is an infinitesimal element of velocity in an element of
time assumed constant, not an instantaneous rate of change of velocity. It is therefore not to
be confused with Newton’s acceleration. Thus in a time dt, velocity v is as dr, and solicitation
is as dv or ddr. As Leibniz tells Burcher de Volder, “solicitations are as dx, velocities are as x
[= ∫dx], and forces as xx or ∫xdx”.32 We shall return to this point later.
Now we are in a position to examine Leibniz’s published solution in the Tentamen for the
centripetal force toward the centre of a body orbiting that centre in an elliptical orbit. Again,
this force for him will be a solicitation, a “dead force”, and as such will be an element of a
conatus proportional to dv and thus ddr. So as in the case of the circle above, he needs to
find an expression for ddr based on the properties of the geometric figure, in this case the
ellipse. Following Newton’s lead, he uses the properties of the ellipse. But he will use them
to set up a differential equation—indeed, the very first differential equation ever published—
31 Tentamen de Motuum Coelestium Causis, Leibniz’s major work on cosmology, in which he re-derives Newton’s results in accordance with his own physico-mathematical principles, was published in the Acta of Leipzig in February 1689; I follow the English translation given by Bertoloni Meli, An Essay on the Causes of Celestial Motions (1993, 126-142; 131). 32 Leibniz to De Volder, GP II 154, 156; quoted by Bertolini Meli 1993, 88.
Leibniz’s Syncategorematic Infinitesimals R. T. W. Arthur
42
rather than proceeding in the classical manner with ratios of quantities. Thus, for an ellipse
with distance from one focus r, minor axis BE = e, major axis AΩ = q, eccentricity e, and latus
rectum XW = a, we have the relations
e2 = q2 – b2 (4.9)
b2 = aq (4.10)
To simplify, Leibniz makes the substitution
p = 2r – q (4.11)
and obtains from an analysis of the geometry of the figure the proportion
(aθ/r):dr :: b:√(e2 – p2) (4.12)
where aθ is twice the area swept out by the radius in the element of time θ = dt, with the
constant a arbitrarily set equal to the latus rectum.33 Leibniz rewrites (4.12) as
br dr = aθ √(e2 – p2) (4.13)
“which is the differential equation” (Meli, 137). Using the fact that with p = 2r – q, we have
dp = 2dr, so that
d√(e2 – p2) = –2pdp/2√(e2 – p2) = –2pdr/√(e2 – p2),
and remembering that aθ is constant, Leibniz differentiates (4.13), to obtain:
b dr dr + br ddr = –2paθ dr /√(e2 – p2) (4.14)
“Eliminating dr from these two equations [4.13 and 4.14]” (Meli, 137), Leibniz obtains:
b a2θ2(e2 – p2)/b2r2 + br ddr = –2p a2θ2/br (4.15)
⇒ ddr = a2θ2(–e2 + p2 – 2pr)/b2r3 (4.16)
33 Here Leibniz makes a trivial slip, as pointed out by Bertoloni Meli (1993, 118). In order for aθ to be twice the area of the elementary triangle, a needs to be half the latus rectum. Thus XW = 2a, and b2 = 2aq, introducing an error of a factor of 2 in the second term of equation (4.19).
Leibniz’s Syncategorematic Infinitesimals R. T. W. Arthur
43
Now, using the relations (4.9) e2 = q2 – b2 and (4.11) p = 2r –q, we can calculate that
–e2 + p2 – 2pr = b2 – 2qr, (4.17)
thus giving
ddr = a2θ2(b2 – 2qr)/b2r3 (4.18)
or, since by (4.10) b2 = aq,
ddr = a2θ2/r3 – 2aθ2/r2 (4.19)
Here Leibniz identifies the first term as twice the centrifugal conatus, and the second as the
solicitation due to gravity, in agreement with Newton’s inverse square law. Taking into
account Leibniz’s slip in not taking a to be the semi-latus rectum (see footnote 37), so that b2
should in fact be 2aq, the result should be
ddr = a2θ2/r3 – aθ2/r2 (4.20)
This derivation can be compared with a modern derivation using differentials as follows.
In spherical co-ordinates (r, α) the equation of the ellipse is
r = a/(1 + e·cosα) (4.21)
where a is the semi-latus rectum and e the eccentricity. Differentiating and rearranging, we
obtain
dr = ea·sinα·dα/(1 + e cosα)2
= er2·sinα·dα/a (4.22)
Now by Kepler’s Law (cf. (3.15) and (3.16)), we have r2dα = adt, giving
dr = e·sinα·dt
⇒ d2r = e·cosα·dα·dt = ae·cosα·dt2/r2 (4.23)
Leibniz’s Syncategorematic Infinitesimals R. T. W. Arthur
44
But by (4.20), e·cosα = a/r – 1, giving
d2r = (a2/r3 – a/r2)·dt2 (4.24)
agreeing with Leibniz’s result (4.19), as corrected in (4.20).
Eschewing differentials and using only derivatives, an equivalent result can be derived as
r'' = (a2/r3 – a/r2) (4.25)
5. SECOND-ORDER DIFFERENTIALS
This addresses the mathematical physics. But a crucial question remains: is Leibniz entitled to
his second-order differentials, given his understanding of first order differentials as fictions?
And can SIA achieve a similar result? In this section I shall first consider whether second-order
differentials are definable on the syncategorematic interpretation of infinitesimals, and then
whether they are definable in SIA. Then I will proceed to determine whether Leibniz’s
calculations above are justifiable on such a syncategorematic interpretation, and whether a
similar calculation might be achieved using SIA. These considerations will motivate a revised
treatment of differentials in SIA, as well as lead to some reflections on the relationship
between Leibniz’s infinitesimals and the infinitesimals of SIA.
The question of Leibniz’s entitlement to second-order differentials has been subjected to
a lucid examination by Henk Bos in his (1974-75). As Bos demonstrates, Leibniz’s own attempt
to justify rules involving second-order differentials in Cum prodiisset is compromised by his
desire to produce differential equations that do not depend on the “progression of the
variables” chosen, i.e. on the specification of an independent variable. In setting up his
general scenario, Leibniz lets the variables dx and ddx vary independently, the only condition
being that when dx → 0, also ddx → 0. But as Bos explains, he also needs ddx to be infinitely
small with respect to dx, i.e. for ddx/dx → 0 as dx → 0. The condition that guarantees this in
general is that dx be constant, so that ddx = 0.
Leibniz’s Syncategorematic Infinitesimals R. T. W. Arthur
45
Y2
T1
As Bos points out, Leibniz’s assumption that dx and ddx vary independently is visible in
his attempt to justify the rules for second-order differentials by assuming two constant lines
of different lengths, (d)x and (dd)x —which he also writes 2(d)x— and two variable lines (d)y
and (dd)y, which are such that
(d)y:(d)x = dy:dx as before, and (5.1)
(dd)y: (dd)x = (dy + ddy):(dx + ddx) (5.2)
He then assumed that in the limit as dx → 0, (dd)y = (d)y. As Bos observes, however, this is
the case only if in that limit (dd)x = (d)x. This is equivalent to assuming x as independent
variable, i.e. an arithmetic progression in x, so that dx is constant and ddx = 0. On this
assumption, however, as Bos goes on to show, Leibniz’s formula may be successfully derived
along the lines he proposed, consistently with the syncategorematic interpretation described
above.
Figure 9: Second-order Differentials34
A (d)x 2(d)x
Ω1
X1 Y1
Ω2
X2
X3 Y3
34 This is a slightly simplified version of Leibniz’s figure, with the abscissas drawn vertically as was his custom, and with his 1X, 1Y, etc., written as X1, Y1, etc., as before.
Leibniz’s Syncategorematic Infinitesimals R. T. W. Arthur
46
Here dx = X2 – X1, ddx = X3 – X2, y = X1Y1, dy = X2Y2 – X1Y1, and ddy = X3Y3 – X2Y2. In his
construction Leibniz takes the varying line (d)y defined by (5.1) to be equal to a second
varying line X1ω1, in such a way that, in the limit where Y1 and Y2 coincide, when it becomes
X1Ω1, it bears to the fixed line (d)x the ratio of the ordinate X1Y1 to the subtangent X1T. This
accords with his definition (2.2) of dy:dx as being equal in the limit to (d)y:(d)x = y:σ. “Since
this can be done wherever the point Y1 is assumed on the curve,” he writes, “in this way a
curve ΩΩ will be produced which is the differentrix of the curve YY” (Leibniz 1846, 49). A
similar construction is then given for 2(d)y in terms of 2(d)x in terms of a second differentrix.
Leibniz uses the same approach in his published reply to Nieuwentijt in 1695, noting that a
construction of such a differentrix is always possible.35
The particular formula that Leibniz then proceeds to demonstrate in Cum prodiisset is
that for the second-order differential ddy given that ay = xv, where a is a constant to
preserve the homogeneity of the equation. For finite differences we have
a(y + dy) = (x + dx)(v + dv) (5.3)
and
a(y + 2dy + ddy) = (x + 2dx + ddx)(v + 2dv + ddv) (5.4)
giving
ady = xdv + vdx + dxdv (5.5)
and, after some elementary algebraic manipulation,
a·ddy = x·ddv + v·ddx + 2 dv·dx + 2dv·ddx + 2dx·ddv + ddx·ddv (5.6)
Since dx is constant, and ddx = 0, we may drop the terms in ddx. Dividing by dx2 gives
35 See Leibniz’s reply to Nieuwentijt in his (1695), and Bos’s lucid account: “Obviously, the procedure can be repeated again, by which LEIBNIZ has shown that finite line-variables can be given proportional to differentials of any order” (Bos 1974-75, 65).
Leibniz’s Syncategorematic Infinitesimals R. T. W. Arthur
47
a·ddy/dx2 = x·ddv/dx2 + 2 dv/dx + 2dx·ddv/dx2 (5.7)
This is still an equation in variable quantities that may be made as small as desired. But in
order to “transpose the case as far as possible to quantities that never vanish”, we need to
define (d)v analogously to the definition of (d)y in (5.1):
(d)v:(d)x = dv:dx (5.8)
In place of Leibniz’s definition (5.2) and an analogous one for (dd)v, Bos lets
(dd)y:[(d)x]2 = ddy:dx2 (5.9)
(dd)v:[(d)x]2 = ddv:dx2 (5.10)
Assume further that these ratios all have an interpretation when dx = 0. Then each of (5.8),
(5.9) and (5.10) can be substituted for its counterpart involving dx in (5.7), yielding
a(dd)y/(d)x2 = x·(dd)v/(d)x2 + 2(d)v/(d)x + 2dx·(dd)v/(d)x2 (5.11)
By hypothesis, this formula remains interpretable when dx = 0, when the last term vanishes.
So the Law of Continuity asserts that this limiting case may also be included in the general
reasoning: dv/dx can be substituted for (d)v/(d)x etc. in the resulting formulas even for the
case where dx = 0, with dy, dx, dv, ddy, in this case interpreted as fictions. Therefore
a ddy/dx2 = x·ddv/dx2 + 2 dv/dx (5.12)
or
a ddy = x·ddv + 2 dv·dx QED. (5.13)
This is the formula that Leibniz was aiming to derive, achieved by Bos (1974-75, 61-62)
using Leibniz’s method together with his own creative emendations. Although it is only a
particular result, the point is that the assumptions it depends upon are quite general: an
analogous construction is possible for any variable y that is differentiable in the
neighbourhood of the point in question, i.e. is a smooth function of x. The particular result
Leibniz’s Syncategorematic Infinitesimals R. T. W. Arthur
48
comes with the caveat that it is only valid for x as independent variable, and is not a general
formula relating the second-order differentials ddy and ddv. This can be made more
perspicuous by adopting the notation used by Hermann Grassmann in his Ausdehnungslehre of
1862 (2000, 249ff), in which (5.13) would be written
a dx2y = x· dx
2v + 2 dxv·dx (5.14)
Bos is of the opinion that the need to specify second derivatives with respect to an inde-
pendent variable “needlessly restricts the generality of the differential calculus, as it imposes
the choice of a special progression of the variables” (66). But by his own arguments, on the
one hand second-order differentials are not well defined without this restriction; and, on the
other, second-order differentials may in any case often be circumvented. As Bos explains,
Leibniz showed, in his treatment of the evolute as envelope of the family of the normals to a
curve, how higher order differentials can be eliminated by calculating the differential
equation of the curve, yielding formulas for the radius of curvature that are independent of
the progression of the variables (Bos 1974-75, 41). Bos suggests that the fact that this
expedient, as well as Leibniz’s own attempts to found the calculus (by what I have called his
“finite surrogate” methods), led him to introduce differential quotients, defined as ratios of
finite quantities that remain finite in the limit as dx → 0, took him a long way towards the
modern definitions of function and derivative of a function (66). This is perfectly true; but it
is no criticism of Leibniz’s founding of the calculus in terms of infinitesimals when these are
understood as fictions under those very conditions: not as actually infinitely small quantities,
but as finite quantities that may be made as small as desired, and which may be treated in
the limit as if they are infinitely small. Thus the equation Leibniz derives at the end of Cum
prodiisset relating second-order differentials “will hold when the points X1 and X2 coincide”,
he writes, “if, by a certain fiction, we suppose dx, ddx, dv, ddv, dy, and ddy to remain even
Leibniz’s Syncategorematic Infinitesimals R. T. W. Arthur
49
when evanescent, as if they are infinitely small quantities (and there is no danger in this,
since the matter can always be referred back to assignable quantities)” (Leibniz 1846, 50).
This establishes Leibniz’s right to use second-order differentials under that fiction,
provided an independent variable is identified.
Now let us turn to the question of higher order differentials in Smooth Infinitesimal
Analysis. I should emphasize at the outset that SIA does not actually need them; it is perfectly
able to derive a formula such as (5.12) under the assumption that y and v are functions of x as
an independent variable, with ddy/dx2 understood as the second derivative of y with respect
to x, and similarly with ddv/dx2 and dv/dx.
Thus if ay = xv, with a constant, by the product rule (3.11), y´(x) = f´(x)g(x) + g´(x)f(x),
ay´(x) = xv´(x) + v(x) (5.15)
Applying the same rule to take the derivative of this equation, we have
ay"(x) = xv"(x) + 2v´(x) (5.16)
which is the analogue of (5.12) in derivative form. By the same token, SIA is easily able to
derive (4.25), r'' = (a2/r3 – a/r2), the formula for acceleration in an elliptical orbit.
Those calculations, however, involve derivatives, not differentials. The question we are
treating is whether Leibniz’s infinitesimals are successfully modelled by the infinitesimals of
SIA, and his derivations depend on the use of second-order differentials. Are these definable
in SIA? Indeed they are, and I am indebted to John Bell for providing me with the following
account of how this should be done.36
36 Bell’s definition of second-order differentials here differs from that given in his (1998, 90-91), where they are defined as nilcube infinitesimals, ones such that ε3 = 0. But as I pointed out in an earlier version of this paper, this makes the decision of what type of infinitesimal to adopt (nilsquare or nilcube, etc.) depend on their applicability to the problem at hand; moreover, all the principles of SIA that we have depended upon above depend critically on the nilsquare property, which fails if the
Leibniz’s Syncategorematic Infinitesimals R. T. W. Arthur
50
The basic idea is that, for a map h: R → R, the nth order differential dnh should be taken
to be the “nth order infinitesimal part” of the difference in the value of h(x) when x is
replaced by x + ε1 + ε2 + … + εn, with ε1, …, εn ∈ Δ. This difference is
h(x + ε1 + ε2 + … + εn) – h(x)
= (ε1 + … + εn)h'(x) + (ε1ε2 + … + εn-1εn )h''(x) + … + ε1…εn h(n)(x). (5.17)
The “nth order infinitesimal part” of the difference is the last term on the r.h.s., namely
ε1…εn h(n)(x). Accordingly, for n ≥ 1 we define
dnh: R × Δn → Δ
by
dnh (x, ε1, …, εn) = ε1…εn h(n)(x). (5.18)
For example
dh(x, ε) = εh'(x), d2h(x, ε, η) = ε·η·h''(x) (5.19)
Using this definition, we may derive the formula for the second derivative of a product. Given
f, g: R → R, let h = f.g. Then for ε, η ∈ Δ, we have
d2h(x, ε, η) = εη(fg)'' = εη(f'g + fg')' = εη(f''g + 2f'g'+ fg'')
= εηf''g + 2εf'ηg'+ εηfg''
= g·d2f + 2df·dg + f·d2g (5.20)
From this it is easy to derive Leibniz’s formula (5.13): if ay = xv, then a·ddy = x·ddv + 2 dv·dx.
For in this case f(x) = x, g(x) = v, so that f' = 1, f'' = 0, df = dx = const. and d2f = 0 , giving
a·d2y = d2(xv) = xd2v + 2dx·dv (5.21)
infinitesimals are nilcube. Bell has generously offered the present account (private communication) to circumvent such difficulties.
Leibniz’s Syncategorematic Infinitesimals R. T. W. Arthur
51
Now let us proceed to the question of whether the formulas Leibniz derived for orbiting
bodies, such as (4.7) and (4.20), are in accordance with these foundations. Taking Leibniz’s
infinitesimals first, we have seen that second-order differentials are well defined under his
syncategorematic interpretation, provided an independent variable is identified. That is the
case here. We are assuming, as both Leibniz and Newton do, that time may be taken as
independent variable, so that dt is constant, and ddt = 0. Under this assumption one may
derive legitimate formulas involving second-order differentials such as ddr, like those derived
by Leibniz above. In those formulas, ddr implicitly depends on a choice of t as the
independent variable. But in such a case, with dr = r´dt, we have
ddr = dr´dt + r'ddt
= dr´dt (since ddt = 0)
= d/dt(r´)·(dt)2 = r''·(dt)2 (5.22)
For example, in Leibniz’s chord composition for centrifugal solicitation in a circular motion,
the solicitation ddr is represented by a line element which is traced out with a uniform
velocity w in the time θ = dt. Thus the velocity w is a linear function of the (infinitesimal)
time, and we may set
w = f0 + f1θ (5.23)
where f0 and f1 are functions of r and the transverse velocity v. We may justify this
analogously to the Leibnizian justification of Newton’s Lemma 9 we gave above. First we
expand w in terms of θ:
w = f0 + f1θ + f2θ2 + f3θ3+ f4θ4 + … (5.24)
But since w = 0 when θ = 0, it follows that f0 = 0. Now since QR = w·dt = wθ, we have
ddr = wθ = f1θ2 + f2θ3 + f3θ4+ f4θ5 + … (5.25)
Leibniz’s Syncategorematic Infinitesimals R. T. W. Arthur
52
⇒ ddr/θ2 = f1 + f2θ + f3θ2+ f4θ3 + … (5.26)
Now in the case of the circle, QR = PQ2/SP, and since equal areas of the elementary triangles
PQS are described in equal times, we have PQ·SP = aθ, a being a constant. Therefore
ddr = QR = PQ2/SP = (aθ)2/SP3 (5.27)
⇒ ddr/θ2 = a2/SP3 (5.28)
But since a2/SP3 is finite and non-vanishing when θ → 0, it follows by comparison with
equation (5.26), that f2 , f3, f4, … are all 0. Therefore
ddr = f1θ2 , with w = f1θ (5.29)
Turning now to the problem of reproducing these results using the infinitesimals of
Smooth Infinitesimal Analysis, we are immediately confronted with the difficulty that the
basic principles of SIA as expounded so far, as embodied in the Principle of Microuniformity,
will not countenance time variation of geometric quantities across an infinitesimal interval.
As Whiteside (1966) has shown, if we want the infinitesimal elements of the curve to be
rectilinear, as they are by the Principle of Microstraightness, then they must be second-order
infinitesimals, in contradiction to Microuniformity. Relatedly, if we try to duplicate the
Leibnizian calculation above using nilsquare infinitesimals, this means that any second-order
differential, such as Leibniz’s ddr representing solicitation, is identically zero. This can be
seen as follows. If the moment dt is taken to be a nilsquare infinitesimal, then PQ2 = (vdt)2 =
v2dt2 = 0. Thus
QR = ddr = PQ2/SP = 0 (5.30)
That is, to use Leibniz’s terms, there can be no solicitation along QR. In Newton’s terms, if
the force at the very beginning of the interval is “given as the square of the time inversely”,
i.e. as 1/(SP•QT)2 with SP•QT proportional to dt, then the acceleration in the moment dt is
Leibniz’s Syncategorematic Infinitesimals R. T. W. Arthur
53
undefined, according to SIA. For the nascent triangle SPQ has an area dA = 1/2SP•QT = kdt, so
(SP•QT)2 = 4dA2 = 4k2dt2 = 0. In sum, because ε2 = 0, we are unable to apply SIA to Newton’s
Propositions 4, 6 and 11.
Looking back to Bell’s definition of second-order differentials and his derivation of (5.21)
a·d2y = d2(xv) = xd2v + 2dx·dv
we see that if x is the independent variable, then dv = v'·dx and d2v = v''·(dx)2, so that
a·d2y = (x·v''+ 2)·(dx)2 = 0 (5.31)
In other words, if x is the independent variable, then the infinitesimals ε and η of (5.19) must
differ at most by a constant factor k, so that ε·η = kε2 = 0, and d2h(x, ε, η) is identically zero.
How does this situation bear on Leibniz’s foundation for infinitesimals? For, as Bell
observes (1998, 1-5), Leibniz’s non-punctiform infinitesimals are in many respects analogous
to those of SIA. The Leibnizian polygonal representation of curves is closely related to Bell’s
Principle of Microstraightness: in each case the curve is analyzed as compounded of
infinitesimal rectilinear segments that are in a certain sense fictional parts. But whereas in
SIA the “area deficit” is simply stipulated to be zero (Bell 1998, 8) —that is, it is shown to be
of the order of dx2, where x is the independent variable, and thus rigorously equal to zero— in
Leibniz’s justification of Riemannian integration the Area Deficit is shown to be zero in the
limit by an application of the Archimedean Axiom without any assumption about the
nilpotency of infinitesimals. In fact, using Leibniz’s method it is possible —as we saw in the
above proofs of Lemma 9 and Propositions 4, 6 and 11— to have the second-order differential
ddr proportional to the square of a first order infinitesimal. Because QR = QP2/SP, with QP
first order infinitesimal and SP finite, QR must be second-order infinitesimal to preserve
orders of infinity. This is impossible with nilsquare infinitesimals.
As we saw, Leibniz was able to derive the result (4.7),
Leibniz’s Syncategorematic Infinitesimals R. T. W. Arthur
54
du = ddr ∝ v2/r (5.32)
for the centrifugal solicitation of a body revolving in a circular orbit in an arbitrary
infinitesimal interval of time dt. Here du is an element of velocity, and a simple integration
of it gives a velocity, i.e., on Leibniz’s understanding, increases its order of infinity without
changing its dimension. (In his calculations he tends to leave the dependence on the time
interval tacit, as here with the dependence on dt2.)37 Similarly, two integrations of ddr raise
it two orders of infinity, and yield a finite line r.
We would read Leibniz’s expression, however, as giving us a second derivative with
respect to time,
d2r/dt2 = v2/r = du/dt (5.33)
namely an acceleration, from which the velocity u is obtainable by an integration with
respect to time, and the radius r by two such integrations. So from a modern perspective a
non-zero solicitation in a moment or “timelet” dt is equivalent to an acceleration, contrary to
Leibniz’s understanding. But when Leibniz’s second-order differentials are recognized to be
functions of an independent variable (here, time), the corresponding adjustment to the
understanding of differentials strips him of his grounds for resisting the composition of a
uniform motion from an inertial motion and an accelerated one: what Leibniz conceived (in
(4.5)) as an inertial motion along QR with a velocity w = du in a time dt, is equivalent to a
uniform acceleration du/dt. Thus in motion under a central force, Microstraightness implies
the failure of Microuniformity.
Thus Leibniz is able to uphold the Principle of Microuniformity because of a different
understanding of how physical quantities are integrated in the calculus; but a successful
grounding of second-order differentials through the syncategorematic interpretation leads to
37 See Bos (1974-75, 5-10, 12-35) and Bertoloni Meli (1993, 66-73) for clear expositions of the difference between Leibniz’s understanding of differentiation and integration and the modern conception.
Leibniz’s Syncategorematic Infinitesimals R. T. W. Arthur
55
an abandonment of this position, and with it, the Principle of Microuniformity. SIA, on the
other hand, is able to uphold Microuniformity, à la Nieuwentijt, only by equating the squares
of all first-order infinitesimals to zero, and therewith rejecting the entire apparatus
underlying Newton’s calculation of central forces, from Lemma 9 through to Proposition 6 and
its corollaries and applications —all of which involve time variation of quantities across an
infinitesimal time, and quantities depending on the squares of nascent or evanescent areas
which are, as in the Nieuwentijtian calculus, necessarily equal to zero.
5. CONCLUSION
Despite many points in common, we have seen that Leibniz’s syncategorematic approach to
infinitesimals and the theory of Smooth Infinitesimal Analysis are by no means equivalent.
Indeed, the conceptions of infinitesimal quantities at the heart of each approach are radically
diverse. According to John Bell, “The property of being a nilsquare infinitesimal is an intrinsic
property, in no way dependent on comparisons with other magnitudes or numbers.” (Bell,
1998, 2). Leibniz’s syncategorematic infinitesimals, by contrast, are essentially comparative,
and are defined by an implicit limiting process. The application of the Law of Continuity
presupposes the existence of a limiting value, and also the smoothness of the co-varying
quantities in the neighbourhood of this value. On this approach, the vanishing of infinitesimal
quantities is always a comparative affair, and is grounded on a strictly Archimedean
geometry. It remains to be seen, however, whether this Leibnizian syncategorematic
approach can be set on a foundation as adequate to modern mathematical standards of rigour
as is SIA.
Leibniz’s Syncategorematic Infinitesimals R. T. W. Arthur
56
Acknowledgements
An earlier version of this paper was originally submitted for publication in September 2005 in
a volume on infinitesimals edited by William Harper and Craig Fraser, and remained available
on the internet in the meantime. This explains the fact that references have been made to it
in articles that have already been published, such as Levey 2008, Arthur 2008, Arthur 2009.
But this revised version has, I hope, been much improved by the criticisms and constructive
feedback of Niccolò Guicciardini and John L. Bell; any remaining errors or misunderstandings
are my responsibility alone.
References
Arthur R. T. W. (2006). “The remarkable fecundity of Leibniz's work on infinite series," Annals
of Science, 63(2), 221-225.
Arthur R. T. W. (2008). “Leery Bedfellows: Newton and Leibniz on the Status of
Infinitesimals,” in Infinitesimal Differences: Controversies between Leibniz and his
Contemporaries, ed. Ursula Goldenbaum and Douglas Jesseph, Berlin and New York: De
Gruyter, 2008, 7-30.
Arthur R. T. W. (2009). “Actual Infinitesimals in Leibniz’s Early Thought”, in The Philosophy
of the Young Leibniz, Studia Leibnitiana Sonderhefte 35, ed. Mark Kulstad, Mogens
Laerke and David Snyder, 2009, 11-28.
Bell, J. L. (1998). A Primer of Infinitesimal Analysis. Cambridge: Cambridge University Press.
Bertoloni Meli, Domenico (1993). Equivalence and Priority: Newton versus Leibniz. Oxford,
Clarendon Press.
Bos, H. J. M. (1974-75). “Differentials, Higher-Order Differential and the Derivative in the
Leibnizian Calculus”, Archive for the History of the Exact sciences, 14, 1-90.
Leibniz’s Syncategorematic Infinitesimals R. T. W. Arthur
57
Brackenridge, J. Bruce (1995). The Key to Newton’s Dynamics: the Kepler Problem and the
Principia. Berkeley: University of California Press
Breger, Herbert (2008). “Leibniz’s Calculation with Compendia”, in Infinitesimal Differences:
Controversies between Leibniz and his Contemporaries, ed. Ursula Goldenbaum and
Douglas Jesseph, Berlin and New York: De Gruyter, 2008, 185-198.
Cohen, I. Bernard and George E. Smith (2002). The Cambridge Companion to Newton.
Cambridge: Cambridge University Press.
Gerhardt, C. I. ed. (1849-63). Leibnizens Mathematische Schriften (Berlin and Halle: Asher
and Schmidt; reprint ed. Hildesheim: Olms, 1971), 7 vols; cited by volume and page, e.g.
GM II 316.
Gerhardt, C. I. ed. (1875-90). Der Philosophische Schriften von Gottfried Wilhelm Leibniz
(Berlin: Weidmann; reprint ed. Hildesheim: Olms, 1960), 7 vols; cited by volume and
page, e.g. G II 268.
Grassmann, Hermann (2000). Die Ausdehnungslehre: Vollständig und in strenger Form
bearbeitet (1862) [The Theory of Extension, Thoroughly and Rigorously Treated]. English
translation, 2000, by Lloyd Kannenberg, Extension Theory. American Mathematical
Society.
Gregory of St Vincent (1647). Opus Geometricum quadraturae circuli et sectionum coni, 2
volumes. Antwerp.
Guicciardini, Niccolò (1999). Reading the Principia. Cambridge: Cambridge University Press.
Hellman, Geoffrey (2006). “Mathematical Pluralism: The Case of Smooth Infinitesimal
Analysis", Journal of Philosophical Logic, 35: 621-651.
Leibniz’s Syncategorematic Infinitesimals R. T. W. Arthur
58
Ishiguro, Hidé (1990). Leibniz’s Philosophy of Logic and Language (2nd ed.) Cambridge:
Cambridge University Press.
Jesseph, Douglas (2008). “Truth in Fiction: Origins and Consequences of Leibniz’s Doctrine of
Infinitesimal Magnitudes”, in Infinitesimal Differences: Controversies between Leibniz
and his Contemporaries, ed. Ursula Goldenbaum and Douglas Jesseph, Berlin and New
York: De Gruyter, 2008, 215-233.
Knobloch, Eberhard (2002). “Leibniz’s Rigorous Foundation of Infinitesimal Geometry by
means of Riemannian Sums.” Synthese 133: 59-73.
Leibniz, G. W. (1684). “Nova methodus pro maximis et minimis, itemque tangentibus, quae
nec fractas, nec irrationales quantitates moratur, et singulare pro illis calculi genus.”
Acta Eruditorum, 467-473; GM V 220-226.
Leibniz, G. W. (1695). “Responsio ad nonnullas difficultates a dn. Bernardo Nieuwentijt circa
methodum differentialem seu infinitesimalem motas.” Acta Eruditorum, July, 1695, 310-
316; GM V 320-326.
Leibniz, G. W. (1701). “Cum prodiisset atque increbuisset Analysis mea infinitesimalis …” in
Leibniz 1846, 39-50.
Leibniz, G. W. (1846). Historia et Origo calculi differentialis a G. G. Leibnitzio conscripta
(ed. C. I. Gerhardt). Hanover.
Leibniz, G. W. (1923-): Sämtliche Schriften und Briefe, ed. Akademie der Wissenschaften der
DDR (Darmstadt and Berlin: Akademie-Verlag; cited by series, volume and page, e.g. A VI
2, 229.
Leibniz’s Syncategorematic Infinitesimals R. T. W. Arthur
59
Leibniz, G. W. (1993). De quadratura arithmetica circuli ellipseos et hyperbolae cujus
corollarium est trigonometria sine tabulis. Ed. and commentary by Eberhard Knobloch.
Göttingen: Vandenhoek & Ruprecht.
Leibniz, G. W. (2001). The Labyrinth of the Continuum: Writings on the Continuum Problem,
1672-1686. Ed., sel. & transl. R. T. W. Arthur. New Haven: Yale University Press;
abbreviated LoC with page number.
Levey (2008). “Archimedes, Infinitesimals and the Law of Continuity: On Leibniz’s
Fictionalism”, in The Philosophy of the Young Leibniz, Studia Leibnitiana Sonderhefte
35, ed. Mark Kulstad, Mogens Laerke and David Snyder, 2009, 107-133.
Mancosu, Paolo (1996). Philosophy of Mathematics and Mathematical Practice in the
Seventeenth century. Oxford: Oxford University Press.
Nagel, Fritz (2008). “Nieuwentijt, Leibniz, and Jacob Hermann on Infinitesimals”, in
Infinitesimal Differences: Controversies between Leibniz and his Contemporaries, ed.
Ursula Goldenbaum and Douglas Jesseph, Berlin and New York: De Gruyter, 2008, 199-
214.
Newton, Isaac (1967). The Mathematical Papers of Isaac Newton, Volume 1, 1664-1666. Ed.
Whiteside, D. T. Cambridge: Cambridge University Press.
Newton, Isaac (1999). The Principia: Mathematical Principles of Natural Philosophy. Trans. I.
Bernard Cohen and Anne Whitman. Berkeley: University of California Press.
Nieuwentijt, Bernhard (1695). Analysis Infinitorum, seu Curvilineorum proprietatates ex
polygonuarum natura deductae. Amsterdam: Wolters.
Whiteside, D. T. (1966). “Newtonian Dynamics”. History of Science 5, 104-117.